View Full Version : The NEXT LAST R600 Rumours & Speculation Thread
jimmyjames123
13-Apr-2007, 04:16
If the drivers are in as good shape as at least I've understood they should be, AMD should see quite big wins on the OEM sector due nVidias driver situation, not limited to just Delll.
I doubt that NV's "driver situation" would have anything to do with "big wins". By end of May when R600 is expected to launch, I would expect NV to have solid drivers at a minimum.
Skrying
13-Apr-2007, 04:18
I doubt that NV's "driver situation" would have anything to do with "big wins". By end of May when R600 is expected to launch, I would expect NV to have solid drivers at a minimum.
That really does not appear to be the case.. at all. Drivers are actually a huge part of what an OEM would consider. If the driver is not stable then the OEM is the one hearing about it and that simply costs them money after the purchase, something they very much do not like. It is no secret that Nvidia has had driver troubles as of late in Vista, specifically with the 8800 series, unless something radical happens within a month then I have doubts Nvidia is going to be able to improve the driver drastically.
Rangers
13-Apr-2007, 04:22
320 stream processors eh? Looks like AMD decided to bite Nvidia marketing..
I mean it's probably irrelevant to actual performance, but hey, it's a bigger number..
I'm still holding out hope that R600 is a massive performance increase on 8800..hey it's possible.
So are we going to be hearing stuff on R600 April 23/24, or is it one of those NDA lifts two weeks later meetings we talked about?
jimmyjames123
13-Apr-2007, 04:22
That really does not appear to be the case.. at all. Drivers are actually a huge part of what an OEM would consider. If the driver is not stable then the OEM is the one hearing about it and that simply costs them money after the purchase, something they very much do not like. It is no secret that Nvidia has had driver troubles as of late in Vista, specifically with the 8800 series, unless something radical happens within a month then I have doubts Nvidia is going to be able to improve the driver drastically.
Well OEM's use some common sense too. They've had a relationship with NVIDIA for years and years, and all of a sudden they are going to be concerned about "driver issues" from NV and not ATI, when speaking about brand new architecture and brand new OS? That doesn't make any sense. The R600 has not even launched yet, so how could they judge stability and reliability of a product that hasn't even launched yet?
Now, not to be disrespectfull (but of course it'll be taken that way), but why is it that 99% of people who have these cards sent out to them are almost always 'dense'. Gibbo just seems like some kind of 'average user' who's never really done any benchmarking before in his life. Heck, he's asking people for advice on things to run other than 3dmark... And he seems a little ditzy on troubleshooting. Bit of a 'Got an error, what do I do?' type thing.
It reminds me of when people started receiving x800 samples, one of them didn't even know what the hell 3dmark was! Another put on blank stares when someone suggest the (at the time) hugely popular Aquamark. It just seems like ATi's gone out of their way to find braindeads to test their hardware...which makes no sense because such cards should be going to the elite of the community.
:???:
Skrying
13-Apr-2007, 04:25
Well OEM's use some common sense too. They've had a relationship with NVIDIA for years and years, and all of a sudden they are going to be concerned about "driver issues" from NV and not ATI, when speaking about brand new architecture and brand new OS? That doesn't make any sense.
What? Did you not realize that it was a hypothetical situation? If the R600 drivers are more stable than Nvidia's then of course its going to be a win for AMD, but in the end we do not know and therefore its hypothetical.
jimmyjames123
13-Apr-2007, 04:27
I don't understand what point you are trying to make. Why would OEM's speculate on a hypothetical "whose drivers will be better" type of scenario, especially when one card/lineup hasn't even launched yet? It just sounds silly. Both companies have plenty of bugs to work out when faced with a brand new architecture and brand new OS.
Skrying
13-Apr-2007, 04:28
I don't understand what point you are trying to make. Why would OEM's speculate on a hypothetical "whose drivers will be better" type of scenario? It just sounds silly.
OEM are not, we are. Kaotik stated from what he heard that ATI's drivers were more stable, and therefore it would be a win for them. Your comment was further just a guess by stating that you doubt Nvidia will have any more driver issues by R600 launch. It is entirely hypothetical and we are simply making guesses right now on this front.
jimmyjames123
13-Apr-2007, 04:30
I don't think Kaotik really has any personal experience comparing drivers on R600 vs G80. The only way we will know is when the R600 has been selling for a few weeks and people get to directly compare and constrast with G80.
I believe that both companies are working like hell to fix bugs and glitches on their Vista and XP driver sets. This is just compounded by the fact that the architectures are brand new. I don't expect one or the other company to have any significant advantage over the other in terms of driver stability over the next few months. Both driver teams are pretty good. We will see.
overclocked_enthusiasm
13-Apr-2007, 04:47
Seems to me that AMD is holding off on releasing a GDDR4 2900 XTX until they see what Nvidia decides to do. It also seems to me that 2900 XT is likely slightly better than 8800 GTX (IQ and features) and will be judged "the best" by most reviewers and boutique box makers. Slapping a $399 price point on it will prevent Nvidia from migrating the 8800 GTX down in price to undercut the 2900 XT and thus preserve AMD's "win" with the flagship part.
The tricky part is that this "win" will last only as long as Nvidia decides to not unleash the 8800 Ultra or whatever bests the 2900 XT. At that point, AMD can decide to either field a 80 nm GDDR 4 2900 XTX or if all is going VERY well perhaps a 65 nm part billed as 2950 XTX. It all depends on Nvidia which way AMD will be forced to go. Seems like AMD didn't need to release the XTX to beat the GTX so they are holding it in reserve if needed....or they just got lucky with the XT.
If AMD can't beat the refreshed 8800 GTX or Ultra with a souped up 80 nm GDDR4 2900 XTX why on earth would they release it? In that case AMD will be forced to use the R650 to again top Nvidia. So I can see a scenerio where the 80 nm R600 GDDR4 XTX would be canned in favor (and neccesity) of the 65 nm GDDR 4 R650 when it is ready.
While this arguement works both ways, AMD may finally have the catbird seat for a month or more if 2900 XT has IQ and feature advantages that Nvidia can't touch right now. I like the pricepoint of $399 as it will prompt many to finally switch their rigs to Vista IF the drivers are solid. They are playing the volume game at $399 and make no mistake about it.
Lastly, if AMD can show Barcelona is a beast and RD790 is the shiznick than AMD has maybe finally stopped the bleeding.
{Sniping}Waste
13-Apr-2007, 04:47
Well OEM's use some common sense too. They've had a relationship with NVIDIA for years and years, and all of a sudden they are going to be concerned about "driver issues" from NV and not ATI, when speaking about brand new architecture and brand new OS? That doesn't make any sense. The R600 has not even launched yet, so how could they judge stability and reliability of a product that hasn't even launched yet?
The big OEMs like Dell/HP all ready know about the R600/RV630, and RV610 because they been testing them for months now. They get engineering samples long before the launch of a product. It would not surprise me if Dell/HP/other OEMs had R600 around December to test out.
I am neither wise nor knowledgeable (particularly about R600), but hasn't anyone benchmarked an 8800 with both stock and overclocked RAM? Otherwise, all I've got to go on are sites (http://www.tweaktown.com/articles/983/9) that have OCed the RAM along with the core, which isn't entirely useful when trying to determine bandwidth sensitivity. OCed 8800GTXs are hitting 100GB/s with 2GHz RAM, so someone with a card can definitely answer your question.
---
As for R600XTX's purported 12k 3DM06 score, an upclocked 8800GTX seems (http://www.tweaktown.com/articles/983/3/page_3_benchmarks_3dmark06/index.html) capable (http://www.nordichardware.com/Reviews/?page=4&skrivelse=493) of reaching that #, so that figure's not super-interesting in terms of fleshing out R600's strong points. Still, nice to know it'll at least be competitive.
Texture/Fillrate/ROP's and shader clock frequency plays very important part of the role.
But I remember the days when R300 came out and some people said extra bandwidth is a waste; but later 256bit memory served very well for R300 and made sure R300 was fed all the time.
Extra bandwidth may serve later - based on ATI idea - Thats what I think what ATI has in mind.
But I do agree with you about over-clocking GPU+memory on G80 and see FPS increase, but if you will not over-clock GPU itself - you will less likely see higher FPS.
What is interesting maybe R600 needs extra bandwidth to utilize full potential of the core to scale better based on architecture design of the R600: - maybe R600 is memory hungry beast.
100GB/s bandwidth should be normal for games, but extra bandwidth on top is a bonus.
One last thing I want to add that R600 1GB RAM vs. 768MB RAM G80, will be very interesting when limitation start hitting on G80 because of the amount memory it has.
Geeforcer
13-Apr-2007, 05:02
Seems to me that AMD is holding off on releasing a GDDR4 2900 XTX until they see what Nvidia decides to do. It also seems to me that 2900 XT is likely slightly better than 8800 GTX (IQ and features) and will be judged "the best" by most reviewers and boutique box makers. Slapping a $399 price point on it will prevent Nvidia from migrating the 8800 GTX down in price to undercut the 2900 XT and thus preserve AMD's "win" with the flagship part.
The tricky part is that this "win" will last only as long as Nvidia decides to not unleash the 8800 Ultra or whatever bests the 2900 XT. At that point, AMD can decide to either field a 80 nm GDDR 4 2900 XTX or if all is going VERY well perhaps a 65 nm part billed as 2950 XTX. It all depends on Nvidia which way AMD will be forced to go. Seems like AMD didn't need to release the XTX to beat the GTX so they are holding it in reserve if needed....or they just got lucky with the XT.
If AMD can't beat the refreshed 8800 GTX or Ultra with a souped up 80 nm GDDR4 2900 XTX why on earth would they release it? In that case AMD will be forced to use the R650 to again top Nvidia. So I can see a scenerio where the 80 nm R600 GDDR4 XTX would be canned in favor (and neccesity) of the 65 nm GDDR 4 R650 when it is ready.
While this arguement works both ways, AMD may finally have the catbird seat for a month or more if 2900 XT has IQ and feature advantages that Nvidia can't touch right now. I like the pricepoint of $399 as it will prompt many to finally switch their rigs to Vista IF the drivers are solid. They are playing the volume game at $399 and make no mistake about it.
Lastly, if AMD can show Barcelona is a beast and RD790 is the shiznick than AMD has maybe finally stopped the bleeding.
So in other words, you foresee AMD going from People's Glorious Family Launch to "release one card, (2900 XT) see what happens"? That would be quite an about-face as far as their stated strategy/intentions are concerned.
Also, in order to stop the bleeding AMD has to do more than "show" Barcelona. They need to ship it.
Ailuros
13-Apr-2007, 05:32
24xAA? Hmm. Everybody clap for Tinkerbell. . . .
With recent drivers you can enable up to (relatively useless) 32xS on a single G80. And just to clarify the sample placement in that hybrid mode is for the trash can; at least if you compare it directly to 16xS in terms of quality and the rather huge performance penalty compared to the latter.
As I recently said in a private conversation with a friend (yes more OT) those hybrid modes nowadays are only good for older games and if you're limited to a relatively low resolution. For alpha tests there's Transparency AA available and for shader AA I doubt performance allows any portion of Supersampling in a reasonable resolution. What remains is the added LOD offset due to the SS samples.
Uhmmm I'm not implying that those 24xAA-whatever thingies are for a hybrid mode or anything, yet higher sample densities don't exactly mean anything before one sees what it truly stands for.
epicstruggle
13-Apr-2007, 06:05
So in other words, you foresee AMD going from People's Glorious Family Launch to "release one card, (2900 XT) see what happens"? That would be quite an about-face as far as their stated strategy/intentions are concerned.
Agreed, its interesting to see scenarios change so fast. Launch date cant come fast enough.
overclocked_enthusiasm
13-Apr-2007, 06:27
So in other words, you foresee AMD going from People's Glorious Family Launch to "release one card, (2900 XT) see what happens"? That would be quite an about-face as far as their stated strategy/intentions are concerned.
Also, in order to stop the bleeding AMD has to do more than "show" Barcelona. They need to ship it.
Well, they are losing credibility by the second to be sure. If they don't launch the entire family as they stated then they are full of crap. I am still assuming they will launch all as expected with the exception of the 2900 XTX which is pointless if the 2900 XT is faster than G80.
p.s. You won't find a bigger critic of ATI than I for the past several years so I am NOT defending them or their strategy...especially concerning launch dates.
I am still assuming they will launch all as expected with the exception of the 2900 XTX which is pointless if the 2900 XT is faster than G80.
Why 2900XTX be pointless - it is whole point for ATI to have something a lot faster then G80 in order to recover itself as a company because delay after another delay for the new product release, since they are behind the schedule.
Seems to me that AMD is holding off on releasing a GDDR4 2900 XTX until they see what Nvidia decides to do.
I don't think so, I believe AMD has high-clock R650 65nm X2950XTX for the reserve - just in case if nvidia has a surprise for AMD.
Ye, I think we need to bring out the Psycho music again. :D
US
The psycho music gets boring and unexciting when you hear it 3 times a day for many months.
Russell
13-Apr-2007, 07:18
Well, they are losing credibility by the second to be sure. If they don't launch the entire family as they stated then they are full of crap. I am still assuming they will launch all as expected with the exception of the 2900 XTX which is pointless if the 2900 XT is faster than G80.
p.s. You won't find a bigger critic of ATI than I for the past several years so I am NOT defending them or their strategy...especially concerning launch dates.
Keep in mind that there are rumors of a delay for RV610 and 630.
That really does not appear to be the case.. at all. Drivers are actually a huge part of what an OEM would consider. If the driver is not stable then the OEM is the one hearing about it and that simply costs them money after the purchase, something they very much do not like.
So what's the alternative while witing for the R600 miracle, sell last-gen cards forever?
One last thing I want to add that R600 1GB RAM vs. 768MB RAM G80, will be very interesting when limitation start hitting on G80 because of the amount memory it has.
That will happen in a year or so and G80 will be very old by then.
Twinkie
13-Apr-2007, 07:32
That will happen in a year or so and G80 will be very old by then.
but the problem is, it already feels old..
:smile:
but the problem is, it already feels old..
:smile:
LOL, yes. Freaky.
Geeforcer
13-Apr-2007, 07:43
I don't think so, I believe AMD has high-clock R650 65nm X2950XTX for the reserve - just in case if nvidia has a surprise for AMD.
I have it on very good authority that the X2950XTX card held in reserve is not only real but in addition to a 65nm R650 will also feature 2(!) 32-pipeline R520s that ATI has been patiently holding in reserve since the fall of 2005.
but the problem is, it already feels old..
:smile:
lol - funny, goof-ball :lol: :lol:
That will happen in a year or so and G80 will be very old by then.
Not exactly necessary, if games which will be based on DX10 / 10.1 that will have higher texture quality with demanding more memory, then such as Crysis soon release it will make surprisingly change the whole market view this year.
Not exactly necessary, if games which will be based on DX10 / 10.1 that will have higher texture quality with demanding more memory, then such as Crysis soon release it will make surprisingly change the whole market view this year.
Noone will program a game which will push the G80 to the limit this year. They want to sell games to the masses, not only to the 0.01% hardcore crowd.
The_Wolf_Who_Cried_Boy
13-Apr-2007, 08:32
Link (http://dailytech.com/ATI+Releases+More+R600+Details/article6903.htm)
with a 512-bit memory interface with eight channels
How likely is that?
Having reduced memory access granulation to 32-bits x 4 cycle burst with the R520/580 isn't it odd to be increasing it to 64-bits by 8 cycle bursts with GDDR4, for which you would expect the memory controller would be optimised for? I assume that implies being designed for maximum HDR rendering performance but wouldn't it still be loosing efficiency when matched with GDDR4 relative to GDDR3?
Noone will program a game which will push the G80 to the limit this year. They want to sell games to the masses, not only to the 0.01% hardcore crowd.
If your logic is correct, then I will wait for ATI/AMD R650.... :) and skip R600 series all together!
-----> If a "Law" game industry's will not let me drive 240MPH on freeway then I will not buy Ferrari F50 super sport car.....
I don't think Kaotik really has any personal experience comparing drivers on R600 vs G80. The only way we will know is when the R600 has been selling for a few weeks and people get to directly compare and constrast with G80.
I believe that both companies are working like hell to fix bugs and glitches on their Vista and XP driver sets. This is just compounded by the fact that the architectures are brand new. I don't expect one or the other company to have any significant advantage over the other in terms of driver stability over the next few months. Both driver teams are pretty good. We will see.
If I'd have personal experience, I wouldn't say "For what I've heard", would I?
You need to read more closely what others say before commenting in such way.
What I said is based on 1) State of drivers for DX9 cards vs nVidias, 2) State of nV 8800's drivers 3) What I've heard and read about ATIs driver situation with R6xx
icecold1983
13-Apr-2007, 09:06
Noone will program a game which will push the G80 to the limit this year. They want to sell games to the masses, not only to the 0.01% hardcore crowd.
you dont think crysis will push a g80 to its limits?
you dont think crysis will push a g80 to its limits?
Not in a sense that it will become unenjoyable or limited in any sense, even if it doesn't give you 50+ fps.
Not in a sense that it will become unenjoyable or limited in any sense, even if it doesn't give you 50+ fps.
It will be playable to a certain extent!
you dont think crysis will push a g80 to its limits?
Some people in this thread http://forum.beyond3d.com/showthread.php?t=39782 seem to think that in its current relatively unoptimized build Crysis is getting 20-25 fps on a single 8800 GTX @1920x1200.
That does not sound like 8800 GTX will be pushed to its limits.
Some people in this thread http://forum.beyond3d.com/showthread.php?t=39782 seem to think that in its current relatively unoptimized build Crysis is getting 20-25 fps on a single 8800 GTX @1920x1200.
That does not sound like 8800 GTX will be pushed to its limits.
Optimize or unoptimized is completely different vs. when you change a game settings which demand more graphic memory....
icecold1983
13-Apr-2007, 09:42
Some people in this thread http://forum.beyond3d.com/showthread.php?t=39782 seem to think that in its current relatively unoptimized build Crysis is getting 20-25 fps on a single 8800 GTX @1920x1200.
That does not sound like 8800 GTX will be pushed to its limits.
people who were at the show also say it wasnt running at 1920 x 1200. and that there wasnt aa or af. and the fps was far below 20 a large portion of the time.
Optimize or unoptimized is completely different vs. when you change a game settings which demand more graphic memory....
Which would be what for example? The chip will be limited elsewhere before it becomes bandwidth limited usually.
Their could be another theory about G80 "GF8800GTX 768MB"
It may have enough memory in certain or a lot of situations, but if Crysis allows G80 to drive up to 200MPH, but if going up-heel - G80 may have hard-time accelerating at that speed, because it is limited on horsepower.
Now you sound confused :lol: Are you talking about the bandwidth or just the amount of memory?
Which would be what for example? The chip will be limited elsewhere before it becomes bandwidth limited usually.
right now don't have nothing yet in order to stress G80.
Now you sound confused :lol: Are you talking about the bandwidth or just the amount of memory?
It is not good example but Which video card will run better? 4 pipeline Radeon 9550 with 256MB -or- 8 pipeline Radeon 9800 with 128MB memory.
Their could be another theory about G80 "GF8800GTX 768MB"
It may have enough memory in certain or a lot of situations, but if Crysis allows G80 to drive up to 200MPH, but if going up-heel - G80 may have hard-time accelerating at that speed, because it is limited on horsepower.
I think he means that Crysis's ultra-high textures will eat up all the 768MB of video ram 8800 GTX has, while 1GB cards would not suffer from such an occurrance.
Russell
13-Apr-2007, 10:05
It is not good example but Which video card will run better? 4 pipeline Radeon 9550 with 256MB -or- 8 pipeline Radeon 9800 with 128MB memory.
9550 duh. I thought everybody knew that. More memory = faster.
Crysis allows G80 to drive up to 200MPH = amount of memory G80 has.
going up-heel - G80 may have hard-time accelerating at that speed = Texture/Fillrate/ROP's and shader clock frequency plays very important part of the role
because it is limited on horsepower = amount memory bandwidth
nicolasb
13-Apr-2007, 10:57
I can't be bothered posting all the links :cool: but according to FUDzilla, the 2900XTX product has now been definitely delayed until at least June - cards showing up at Computex and possibly for sale later in the month. XT still scheduled for May and with a $400 price tag. XT performance comparable to (or slightly better than) 8800GTX.
Silent_Buddha
13-Apr-2007, 11:20
Like I said, it will only force themselves into a price war with a company in a better financial state. It's a gamble that's almost certainly going to backfire and quite likely to put them out of business before they can see to its completion.
Wish I could read and respond to posts more than once a day.
Anyways, this isn't the first time AMD has cut margins drastically while having a better product in order to gain marketshare. In fact, they were in much worse condition before they launched the Athon XP than they are now. And they were arguably in a worse position prior to the launch of the Athon 64 and Opteron than they are now.
At worst, AMD will find themselves in the position they were in when they launched the Athlon XP. They'll have to do far far far worse to go back to the days of the K-6. and their failed 586 attempt.
Athon XP when it first launched was faster than Intel's competing chip, yet AMD still undercut Intel by a drastic amount while maintaining virtually no margins on the chip. All this to increase marketshare and more importantly mindshare. Up to that point AMD chips hadn't had a good competitive chip since the 486 days.
Intel of course responded by improving the performance of the P4 with a higher FSB and memory speeds thus eroding the performance edge.
It was until Opteron and the Athon 64 that AMD was able to maintain high margins with their chips. And actually post consistant profits rather than posting consistant losses.
From the looks of things, AMD is going back to what worked for them in the past. They are currently behind both Intel and Nvidia. Both in marketshare AND in mindshare.
If...and that's a VERY big IF the price point is true on R600, it could be in indication that they are trying to replicate what got the to where they were with the A64.
Mind you this is all theory based on a rumor. But it isn't like AMD hasn't done this sort of thing in the past. And I would imagine not the last time they will do it in the future either.
Regards,
SB
In fact, they were in much worse condition before they launched the Athon XP than they are now. And they were arguably in a worse position prior to the launch of the Athon 64 and Opteron than they are now.
Not really, since their expenses back then were a fraction of what they have now.
Also, the $400 XT in May/June won't be cheap by any means. The G80 GTX will also be much cheaper by then, there is simply no possible way to undercut nV on prices this way. nV simply have loads of pricing headroom already now, let alone in June.
Not really, since their expenses back then were a fraction of what they have now.
Really? By that metric Intel must be in a world of hurt... :lol: ;)
nexus_alpha
13-Apr-2007, 12:30
Not really, since their expenses back then were a fraction of what they have now.
Also, the $400 XT in May/June won't be cheap by any means. The G80 GTX will also be much cheaper by then, there is simply no possible way to undercut nV on prices this way. nV simply have loads of pricing headroom already now, let alone in June.
The cheapest GTX I can find is $548 I really doubt the price would drop by more than a hundred dollars in such a short period of time.
Please keep the financial discussions in the appropriate forum, which is 3D Graphics Companies, Industry, & Misc. For example, in this thread (http://forum.beyond3d.com/showthread.php?t=40455), especially so since I figure the article is arguably very related to the discussion...
trinibwoy
13-Apr-2007, 12:40
The cheapest GTX I can find is $548 I really doubt the price would drop by more than a hundred dollars in such a short period of time.
Yeah that would be a big drop in a short time. I expect to see a bunch of rebate offers at first. This is an agressive (or is desperate?) move from AMD though - a GTX beater for $400? Sign me up!
But Nvidia has a lot of free high-end margin from the past few months to mitigate any hit from a future price drop. Can't wait to see their response, if they are able to respond that is.
Yeah that would be a big drop in a short time. I expect to see a bunch of rebate offers at first. This is an agressive (or is desperate?) move from AMD though - a GTX beater for $400? Sign me up!
I'm guessing desperate, if it's true that is. Why would they need to sell even a cutdown R600 for 400$ if it's got more features + performance then the 8800 GTX ?
But Nvidia has a lot of free high-end margin from the past few months to mitigate any hit from a future price drop. Can't wait to see their response, if they are able to respond that is.
The 8800 GTX has been on the market for quite some time now. The price is afaik pretty much the same as when it was released, at least in Sweden. So i'm guessing that they have huge margins on the GTX atm.
trinibwoy
13-Apr-2007, 12:55
I'm guessing desperate. Why would they need to sell even a cutdown R600 for 400$ if it's got more features + performance then the 8800 GTX ?
To claw back market share ASAP? I guess that would fall into the desperate category. But they must believe that their cost structure is much better than Nvidia's for them to think that Nvidia wont just slash to match. And in the end they both lose. I guess dragging industry margins down in order to move product is something they are willing to do as someone pointed out with the AthlonXP. Same thing is happening now with the X2's I guess.
Although if they can cut prices and offer a more compelling product it could be a good move. Nvidia can cut prices to match but they can't do anything in the short-term about a lagging feature-set or performance.
Silent_Buddha
13-Apr-2007, 12:57
Not really, since their expenses back then were a fraction of what they have now.
Also, the $400 XT in May/June won't be cheap by any means. The G80 GTX will also be much cheaper by then, there is simply no possible way to undercut nV on prices this way. nV simply have loads of pricing headroom already now, let alone in June.
Aye, their expenses are higher now but due to maintaining demand and marketshare due to cutting margins they also have significantly higher cash flow than they did back then. Demand would have to take a nosedive and their marketshare would have to fall by half before they found themselves back in a situation similar to what they were in back then. [speaking of the 586, K6, K6-2 days]
True, assuming that HD 2900 XT is similar to or faster than 8800 GTX, Nvidia won't have any choice but to drop prices to a similar level. However, I doubt they will undercut the 2900 by much if at all. Their cost to manufacture should be relatively similar between the two cards. 8 PCB layes vs 6 layers (?) for 8800. 512 mb vs 768 mb for 8800. 1 chip (ati) vs 2 chips (nv).
Also, Nvidia doesn't have a history of cutting into margins like AMD does. Even with their 5xxx series they weren't willing to cut margins to virtually nothing. AMD on the other hand is no stranger to doing this if it fits with their long term plans.
However, if both are the same price, Nvidia will still sell more hardware as there is just more NV fans than there are ATI fans.
-----
Myself, at this point the thought of 24x AA has me drooling at the possibilities.
-----
Considering HD 2900 XTX has been shipping to OEMs for a while now. I'm expecting a delay in the XTX, RV610 and RV630 being due to unexpectedly high demand from OEMs and AMD not being able to stock enough for retail availability while still meeting demand from OEMs.
What I'm wondering is, if OEM's will be required to hold sales of systems featuring HD 2900 XTX, RV610 and RV630 until such time as ATI releases them into the retail channel.
Additionally, while I still think it's a LOW probability, I still can't shake the feeling that HD 2900 XTX - RETAIL is going to be 65 nm R650. While HD 2900 XTX - OEM is going to initially be 80nm R600 with a migration at 65 nm R650 sometime in the future.
After all, OEMs are not adverse to hot running parts requiring excessive cooling. Especially if they also come with a price break. Just look at the original 7900 Quad SLI system that Dell was offering. NOT suitable for Retail, but that didn't bother Dell very much.
Also, I would NOT be surprised if Dell had an order in for a rather large number of R600's. After all, they were the largest OEM purchasers of R420 I believe.
Apple might not ship nearly as many computers as Dell or HP, however, they generally tend to install one graphics companies cards from top to bottom and have a history of generally favoring ATI over Nvidia.
Regards,
SB
Silent_Buddha
13-Apr-2007, 13:00
One more rather important insight.
IF...IF...IF AMD actually is selling a large volume of R6xx products to OEMs.
That would mean they have lost virtually no money by delaying R600 [edit - from the Feb./March timeframe]. It would just mean that instead of making money in the retail channel they are currently making money from OEMs.
IF...IF...IF this is true. That would mean AMD knows how to run a business better than the majority of the forum readers here. :wink: :smile: :grin: :lol: :wink:
Regards,
Croaker
To claw back market share ASAP? I guess that would fall into the desperate category.
Is that really done in the 400$ price bracket ? Especially with much lower priced cards being released very soon.
But they must believe that their cost structure is much better than Nvidia's for them to think that Nvidia wont just slash to match. And in the end they both lose. I guess dragging industry margins down in order to move product is something they are willing to do as someone pointed out with the AthlonXP. Same thing is happening now with the X2's I guess.
Perhaps, but look at Nvidia and the NV30 series. They didn't lower the prices that much and they were late and had a bad product too.
The X2 is dropping in price but it's not necessarily priced that low compared to the performance. Definitely not if you consider the overclocking possibilities with the C2D. Imo, Intel doesn't need to lower it's prices even with AMD's latest price drops.
Although if they can cut prices and offer a more compelling product it could be a good move. Nvidia can cut prices to match but they can't do anything in the short-term about a lagging feature-set or performance.
True. I just don't see the reason why they would sell a great performing chip/card with loads of features for a much lower price then the competition. Especially not if you're in a somewhat more troublesome financial situation then the competition.
trinibwoy
13-Apr-2007, 13:16
IF...IF...IF this is true. That would mean AMD knows how to run a business better than the majority of the forum readers here. :wink: :smile: :grin: :lol: :wink:
Finance 101: $$$ six months ago is worth > $$$ today. Based on your logic, if they go out of business for a year and then get a year's worth of revenue a year from now that would be good business. Believe me, this is not the situation AMD would have chosen. And you are talking about OEM orders for mid-range parts - AFAIK they have never been the focus of any discussions on R600 delays.
Perhaps, but look at Nvidia and the NV30 series. They didn't lower the prices that much and they were late and had a bad product too.
I just don't see the reason why they would sell a great performing chip/card with loads of features for much lower price then the competition. Especially not if you're in a somewhat more troublesome financial situation then the competition.
This is part of the reason why the GeForce FX series is considered the greatest failure of all time. If they had cut down on prices, who knows? They might not have been the best, but at least they'd have been the most popular.
Which brings us to the current situation. I'm pretty sure the R600XT would meet a lukewarm reception at best, if it were to launch at $549. In marketing terms, it has less memory, it arrives 6 months late and it does not have a solid performance edge.
This is the card that would leave reviewers shrugging their shoulders.
Now consider the same card priced at $399. Suddenly all that was working against it turns in favour of it. Less memory doesn't matter, look at the bandwidth on this beast! Arriving later doesn't matter, it has much better Vista drivers and a lot more multimedia features! And just look at the price-performance ratio!
No reviewer will be left shrugging their shoulders in the face of such a deal. You can pretty much expect rave reviews all around. And the consumer will do as he always has - buying the best deal. And that is without a doubt a R600 XT at $399.
This is part of the reason why the GeForce FX series is considered the greatest failure of all time. If they had cut down on prices, who knows? They might not have been the best, but at least they'd have been the most popular.
More popular to some extent. But you also loose mindshare. A Skoda is cheaper then a Audi. But for a good reason. The other problem is that you assume that Ati wouldn't have lowered their prices to compete. That would perhaps have made really Nvidia popular :-) (wouldn't have been good financially though)
Which brings us to the current situation. I'm pretty sure the R600XT would meet a lukewarm reception at best, if it were to launch at $549. In marketing terms, it has less memory, it arrives 6 months late and it does not have a solid performance edge.
This is the card that would leave reviewers shrugging their shoulders.
549$ is to much. Especially if it has 512 Mb of RAM. But the assumption is that it also got an edge on features.
Now consider the same card priced at $399. Suddenly all that was working against it turns in favour of it. Less memory doesn't matter, look at the bandwidth on this beast! Arriving later doesn't matter, it has much better Vista drivers and a lot more multimedia features! And just look at the price-performance ratio!
That's a lot of assumptions. And less memory definitely matters. Just look at the 320 Mb 8800 GTS.
That's a lot of assumptions. And less memory definitely matters. Just look at the 320 Mb 8800 GTS.
I was just trying to predict the reaction of the reviewers ;)
But on the other hand, I do not believe that 8800 GTX will have an upper hand based on the extra 256MB memory. For that to happen, you would have to find a game that uses more video memory than 512MB. I don't think we are at this point yet. I think Hellgate: London or Crysis may change this, but at the same time I'm pretty confident that these games have been developed with the assumption that you have separate settings for 256MB, 512MB and 1024MB. I am highly doubtful that anyone would create a separate quality level for a 768MB card.
Developers will optimize their texture size with usual resolutions in mind. Not extreme ones. So for example, a 2560x1600 framebuffer with 8x MSAA (or 16xQ CSAA, for that matter) and FP16 HDR would require 384MB of memory (2560x1600x8x(8+4)/1024/1000)...
A 4096x4096 shadowmap would take 64MiB, too, for example. (note: I am correctly using MB and MiB here! Neither is a typo! ;) 16:10 ftl?)
It remains to be seen if this does any difference though, but I'm sure we'll know for sure soon enough.
but at the same time I'm pretty confident that these games have been developed with the assumption that you have separate settings for 256MB, 512MB and 1024MB. I am highly doubtful that anyone would create a separate quality level for a 768MB card.
I would guess that the game developers target the cards that are on the market. And at this moment, there's a lot more 768 Mb cards then 1 Gb cards...
Perhaps, but look at Nvidia and the NV30 series. They didn't lower the prices that much and they were late and had a bad product too.
But they could still milk the innovator/market leader image they had back then after the 3dfx demise. Also, they replaced it with the NV35 just a few months later, so it wasn't quite a comparable situation at all. ATI on the other hand has has only problems and delay consistently over the last two years or so, so they're nowhere near the "safe" king-of-the-hill position nV had back then, rather the opposite.
Also, they replaced it with the NV35 just a few months later, so it wasn't quite a comparable situation at all.
I was taking the NV35 into consideration also. Was it faster, had a better featureset then the competition ?
ATI on the other hand has has only problems and delay consistently over the last two years or so, so they're nowhere near the "safe" king-of-the-hill position nV had back then, rather the opposite.
I would argue that there's no safe king-of-the-hill position in the GPU business.
That's a lot of assumptions. And less memory definitely matters. Just look at the 320 Mb 8800 GTS.
Of course memory matters - but where's the point atm where it doesn't matter anymore? Is anything over 512MB "wasted"? Over 768MB? 1GB?
We know that 512MB makes a difference, but how much over it you need to go when it doesn't anymore?
I was taking the NV35 into consideration also. Was it faster, had a better featureset then the competition ?
Of course not, but it was way better than the NV30 and dirt cheap compared to Radeons while being more or less competitive speed-wise, so it was ok from the marketing POV. Hell, I even played Doom3 on my old NV35 when it was freshly released.
I would argue that there's no safe king-of-the-hill position in the GPU business.
Well not permanently, but temporarely at least. They had a good run since the original TNT until then, so the people expected the next Big Thing from them kinda automatically. And noone expected anything useful from ATI prior to R300.
INKster
13-Apr-2007, 14:47
I was just trying to predict the reaction of the reviewers ;)
But on the other hand, I do not believe that 8800 GTX will have an upper hand based on the extra 256MB memory. For that to happen, you would have to find a game that uses more video memory than 512MB. I don't think we are at this point yet. I think Hellgate: London or Crysis may change this, but at the same time I'm pretty confident that these games have been developed with the assumption that you have separate settings for 256MB, 512MB and 1024MB. I am highly doubtful that anyone would create a separate quality level for a 768MB card.
Speaking of Crysis... (http://forums.vr-zone.com/showthread.php?t=143543)
If they are indeed rebuilding the engine and adding new assets, ensuring higher quality shader effects and large resolution textures should be their top priority.
edit
Hey, i'm now in the 1K post club. :D
To claw back market share ASAP? I guess that would fall into the desperate category. But they must believe that their cost structure is much better than Nvidia's for them to think that Nvidia wont just slash to match.
Well, AMD will have the inherent advantage of that 256MB of memory that won't be on the board. I still think that was a major consideration in switching to "XT first" (apparently).
Hey, i'm now in the 1K post club. :D
Well, there goes the neighborhood! :wink:
Congrats. . . new super powers should unlock in the next 24 hrs. . .
Nvidia can cut prices to match but they can't do anything in the short-term about a lagging feature-set or performance.
Or their reputation for Vista.
Jawed
trinibwoy
13-Apr-2007, 15:43
Considering Vista's short life to date and the relatively few people who are actually using it I doubt Nvidia will have any trouble shaking off their Vista foibles if they can deliver solid drivers before R600 launch - people seem to change their opinion on that every two days when another beta leaks :roll: And of course, all that is dependent on the state of AMD's driver offerings.
Considering Vista's short life to date and the relatively few people who are actually using it I doubt Nvidia will have any trouble shaking off their Vista foibles if they can deliver solid drivers before R600 launch - people seem to change their opinion on that every two days when another beta leaks :roll: And of course, all that is dependent on the state of AMD's driver offerings.
Hmm, do you think the OEMs think like that?
Jawed
INKster
13-Apr-2007, 15:59
Considering Vista's short life to date and the relatively few people who are actually using it I doubt Nvidia will have any trouble shaking off their Vista foibles if they can deliver solid drivers before R600 launch - people seem to change their opinion on that every two days when another beta leaks :roll: And of course, all that is dependent on the state of AMD's driver offerings.
But you have to admit it.
Placing the thousands of buyers of a 6 month old (and quite expensive) family of graphics cards in second place in driver development, just to hand priority to software houses who should have had the hardware long before that is a bit hard to swallow. Don't you agree ?
This latest statement (http://www.xbitlabs.com/news/video/display/20070412224213.html) from NV appears to basically regurgitate the words of another official just a few weeks ago, but there's still no sign of a real improvement.
Many of the same bugs are still there, the performance is still lacking, the control panel is still confusing and illogical, the new driver releases are still very far apart and lack anything other than Beta status/non-WHQL, etc, etc.
It's especially baffling to previous longtime buyers of Nvidia hardware, because their drivers used to mean top-notch reliability, and now...
trinibwoy
13-Apr-2007, 16:10
Hmm, do you think the OEMs think like that?
Nope, hence my comment was not referring to them. But we are talking about reputation which I took to imply that people would hold a grudge based on what's happening today. If Nvidia has a compelling product with stable software behind it OEM's are going to push it regardless of what is happening now.
Placing the thousands of buyers of a 6 month old (and quite expensive) family of graphics cards in second place in driver development, just to hand priority to software houses who should have had the hardware long before that is a bit hard to swallow.
Yeah definitely. For the people who are suffering now it's a very pertinent and serious issue. But the guy who buys a card in May isn't going to care about the issues people are experiencing now. But he's sure going to care about the state of things at the time of purchase. I'm just saying that not enough time has passed for the reputation to stick as yet. Lots of people seem to be allowing Nvidia the handicap of new architecture, new OS etc. That handicap will of course disappear upon R600 launch but by that time they could have their drivers in order.
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
The Shaders are divided in four blocks and each has 16 units. We are working to get the number or TMU's and ROPs and GFLOPs. Well, we dont care as long as it runs faster. Stay tuned.
I'm still convinced R600 is a vector architecture, for what it's worth. The implication is R600 is 64 vec5. A 4:1 ALU:TEX ratio then implies 16 TMUs. And, well, 16 ROPs seems likely too.
16 TMUs fits into the ALU:TEX ratio, but it seems awfully low.
Jawed
nicolasb
13-Apr-2007, 16:19
The implication is R600 is 64 vec5.Vec4 + scalar?
pakotlar
13-Apr-2007, 16:29
I have it on very good authority that the X2950XTX card held in reserve is not only real but in addition to a 65nm R650 will also feature 2(!) 32-pipeline R520s that ATI has been patiently holding in reserve since the fall of 2005.
you are evil...
trinibwoy
13-Apr-2007, 16:51
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
I'm still convinced R600 is a vector architecture, for what it's worth. The implication is R600 is 64 vec5. A 4:1 ALU:TEX ratio then implies 16 TMUs. And, well, 16 ROPs seems likely too.
16 TMUs fits into the ALU:TEX ratio, but it seems awfully low.
Jawed
Hmmm blocks of 16 vec-5's are very reminiscent of Xenos. 16 ROPs isn't bad at all at 800Mhz IMO. Especially since they might be considerably beefed up, even more so than G80's.
overclocked_enthusiasm
13-Apr-2007, 16:52
The problem with the assumption that ATI is selling massive quantities of R6xx and RV6xx chips to the OEMs is that AMD just warned bigtime on their Q1 numbers. They forcasted $1.65 Billion (midpoint of guidance) and have recently lowered it to $1.23 Billion. I have a very hard time believing that their was a bunch of R600 products delivered and sold to OEM's in Q1 (Jan - Mar) given those numbers. If they waited till April 1st to begin delivery than all that OEM chatter is possible to be sure.
I still don't see why some are convinced that there is an OEM version of 2900 XTX floating out there that will be used shortly. Why would AMD release a 2900 XTX when the 2900 XT appears to be slightly better, more feature rich and cheaper than 8800 GTX? It makes zero sense to beat the best part on the market with another...especially when it is yours already.
Perhaps if they were binning chips very conservatively initially, then there may be lots of chips that now have the potential to be clocked much higher than the reference 2900 XT. AMD can either hold these chips in reserve for an 80 nm R600 XTX (GDDR3 or GDDR4) in case Nvidia releases something that beats 2900 XT BEFORE R650 is done, continue releasing 2900 XT with increasingly better overclocking headroom or simply allow certain AIB partners to release 2900 XT OC. These are all good options for AMD because they are quick.
Concerning the $399 price I think it was brilliant and caused by the harsh reality of the situation they have put themselves in. They are late, losing market share and have lost tons of mindshare. By moving the price down to $399 they have prevented Nvidia from pushing their one lever they were counting on using against a marginally better 2900 XT...pricing pressure. Nvidia cannot and will not move the 8800 XTX to less than $399 and maintain their gross margins. I do expect them to match AMD's price to keep sales from falling off a cliff. AMD may have finally found the elusive high ground again in terms of having the "best" especially considering the Vista driver situation.
Anyone who thinks the $399 price point is crazy needs to remember their first thought upon hearing that. Mine was "damn I might just upgrade my rig to Vista this year afterall".
trinibwoy
13-Apr-2007, 17:15
Concerning the $399 price I think it was brilliant and caused by the harsh reality of the situation they have put themselves in. They are late, losing market share and have lost tons of mindshare. By moving the price down to $399 they have prevented Nvidia from pushing their one lever they were counting on using against a marginally better 2900 XT...pricing pressure. Nvidia cannot and will not move the 8800 XTX to less than $399 and maintain their gross margins. I do expect them to match AMD's price to keep sales from falling off a cliff. AMD may have finally found the elusive high ground again in terms of having the "best" especially considering the Vista driver situation.
Yeah it seems to be a good move. And I'm sufficiently convinced by geo's point about 512MB of cheap 800Mhz GDDR3 that I don't think margins will be all that bad. Still, an 8-layer PCB and some heavy cooling can't come cheap.
Nvidia would have to be pretty complacent not to have 80nm high-end parts in the pipeline by now. With the current level of 80nm demand any wafer price premium over the 90nm process should be fading quickly. Don't see why they wouldn't have been aiming for a 80nm high-end summer refresh regardless of what AMD did with R600.
Unknown Soldier
13-Apr-2007, 17:20
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
I'm still convinced R600 is a vector architecture, for what it's worth. The implication is R600 is 64 vec5. A 4:1 ALU:TEX ratio then implies 16 TMUs. And, well, 16 ROPs seems likely too.
16 TMUs fits into the ALU:TEX ratio, but it seems awfully low.
Jawed
Maybe it's the XL version? /me ducks
US
http://www.fx57.net/?p=576
If these tests are true, it's very strange to see XT being so much slower than XTX, maybe clocks are consistently lower.
Or, the XT could be what GTS is for GTX.
More numbers, but from vr-zone.
http://www.vr-zone.com/?i=4898
Although the wait for R600 is unbearable but the consoling part is that R600XT stood up well against NVIDIA GeForce 8800 series. Using a Core 2 Extreme QX6700 processor, R600XT scored 12K in 3DMark06, slightly better than a GeForce 8800 GTX. It is benched on the latest 8.361-RC4 drivers, quite an improvement over the older sets of drivers. The core clock of R600XT stood at 750MHz while the memory is at 820MHz. It looks all decent for a US$399 card.
Looks like ATI is getting their drivers in real good shape for the launch. :)
750/820 MHz?
I think the fact that Dell's fastest gaming rigs have XP instead of Vista tells enough of what they think of nVidias drivers at the moment.
I don't doubt for a single second that the card in it will change to R600 as long as it's as fast as GTX, it doesn't even have to be faster, if the drivers are in as good condition as they should be.
Unknown Soldier
13-Apr-2007, 17:38
More numbers, but from vr-zone.
http://www.vr-zone.com/?i=4898
Looks like ATI is getting their drivers in real good shape for the launch. :)
750/820 MHz?
XT getting 12000 at 750 and 820? Wow .. now that is a suprise.
I still think the XT will be 800 core though, memory not too sure.
US
INKster
13-Apr-2007, 17:50
750MHz core... ;)
Thats more than a 8600 GTS, which is a much less complex little beast on a similar 80nm half node.
This R600 XT "thingy" might just make it to my own main rig soon... :D
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
I'm still convinced R600 is a vector architecture, for what it's worth. The implication is R600 is 64 vec5. A 4:1 ALU:TEX ratio then implies 16 TMUs. And, well, 16 ROPs seems likely too.
16 TMUs fits into the ALU:TEX ratio, but it seems awfully low.
Jawed
Comparing that to Xenos which already has 64 vec5 shaders with one quad disabled for redundancy and 16 TMU's (right?) do those specs look likely for R600 considering the years ATi has been working on this design. Unless they are all massively beefed up... XTreme ALUs :p
jimmyjames123
13-Apr-2007, 18:15
If I'd have personal experience, I wouldn't say "For what I've heard", would I?
You need to read more closely what others say before commenting in such way.
What I said is based on 1) State of drivers for DX9 cards vs nVidias, 2) State of nV 8800's drivers 3) What I've heard and read about ATIs driver situation with R6xx
I'm not sure what point you are trying to make. That's pure hearsay. The fact of the matter is that you have no idea. Big deal, neither do most of us until the R600 launches. Give it up already.
A $400 price point sounds pretty good. Sure, their margins will be lower, but the volume will be much bigger. That's only high middle end. The top cards only account for about 1% or less of that volume. As long as they have good long-term contracts with the foundries who make the actual chips, and have an equally good deal on the memory, it makes a lot of sense. And them much more money than selling it at $600.
And it also allows them to bring out a new and faster card barely a few months later. ;)
A $400 price point sounds pretty good. Sure, their margins will be lower, but the volume will be much bigger. That's only high middle end. The top cards only account for about 1% or less of that volume.
The last time I heard them address it in public, AMD was considering that $400+ market to be about 5-8% of the total market. Not sure if that's volume or revenue tho!
Vec4 + scalar?
320 MADs per clock is 64 x 5.
How special functions are handled is, naturally, a mystery.
Jawed
I think the fact that Dell's fastest gaming rigs have XP instead of Vista tells enough of what they think of nVidias drivers at the moment.
Could have something to do with the state of Vista also. Which btw is, imo, rather crappy atm.
A card slightly faster than GTX and costs less than a 640MB GTS? (And not forgetting about the IQ and other things unknown). For the first time in years I have entertained the idea of spending that much on a GPU ..
trinibwoy
13-Apr-2007, 18:43
750MHz core... ;)
Thats more than a 8600 GTS, which is a much less complex little beast on a similar 80nm half node.
This R600 XT "thingy" might just make it to my own main rig soon... :D
You didn't consider voltages. Remember the GTS got up to 1Ghz with a voltage bump :) But all around the 80nm process looks juicy.
The last time I heard them address it in public, AMD was considering that $400+ market to be about 5-8% of the total market. Not sure if that's volume or revenue tho!
Ah, so still a quite small part. I would have guessed it to be in the low teens. Which it might be for a while, when the R600 enters the shops. That's still about ten times as much. So it'll make them a good buck.
You didn't consider voltages. Remember the GTS got up to 1Ghz with a voltage bump :) But all around the 80nm process looks juicy.
That depends on how hot it already is...
The_Skyo
13-Apr-2007, 19:03
So it seems that everyone has jumped on board with the thinking that the XTX isn't coming out soon?
I've been lurking here since september or october waiting for the right time to upgrade, and with my computer having issues with stalker and especially supreme commander, I'm just about sick of waiting. I've been sitting on the money since last summer! I want to get the best in May so I can wait a while before upgrading again. If it's not ready now, any idea when?? They lie so hard if they say it's ready, but they're delaying to launch the whole family, and then the star doesnt even show up!
It's especially baffling to previous longtime buyers of Nvidia hardware, because their drivers used to mean top-notch reliability, and now...
..and now, they have a completely new architecture as opposed to the last few gens which were basically small increments. It's quite a bit more work to do to get that thoroughly tested and bug-free.
As for OEM's, I'm quite sure that they're not in doubt about nV's drivers becoming ok soon given the support they got in the past.
Silent_Buddha
13-Apr-2007, 19:22
Finance 101: $$$ six months ago is worth > $$$ today. Based on your logic, if they go out of business for a year and then get a year's worth of revenue a year from now that would be good business. Believe me, this is not the situation AMD would have chosen. And you are talking about OEM orders for mid-range parts - AFAIK they have never been the focus of any discussions on R600 delays.
I was in fact talking about not only the RV610 and RV630 but also the R600 specifically. OEMs have had their hands on R600 since the beginning of the year. OEMs according to this rumor have been ordering and receiving R600 since at least Feb./March if not earlier.
In particular OEMs have been ordering and receiving HD 2900 XTX. Notice XTX not "just" the XT that will launch first.
Which means AMD is CURRENTLY and has been for a few months making money on R600.
In other words, if the rumors are true that "millions" of R600s are being sold to OEMs that would mean that AMD has lost virtually no money by delaying R600 by approximately 2 (that is TWO) months from Feb./March timeframe to April/May timeframe.
Again, please note the big IF the rumor is true that they are selling "millions" of R600s to OEMs. AMD is selling them to OEMs for a few months now, there is no doubt about that. It's the quantity that is in doubt.
Regards,
SB
Megadrive1988
13-Apr-2007, 19:25
anyone care to speculate what the R650 refresh has over the highend R600 ?
anyone care to speculate what the R650 refresh has over the highend R600 ?
It will be a lot cheaper and a bit cooler, but probably not much faster. It makes more sense to add units than trying to increase the speed significantly. The 65nm process that will be used is about as fast as the 80nm one.
Silent_Buddha
13-Apr-2007, 19:42
Could have something to do with the state of Vista also. Which btw is, imo, rather crappy atm.
Which is odd considering if you have good drivers for your hardware, Vista is actually more stable, more responsive, and only about 1-2 FPS slower. At least with anything I've tried on my rig. Playing videos is significantly improved on Vista also. It's very gratifying to finally be able to move a movie playing in a window from one monitor onto any of the other 3 monitors seemlessly with not hitches or glitches. And same goes for 3D games in a window. Everything is just soo much better in Vista on the same machine.
Then again, I don't currently have an Nvidia card or Creative Labs sound card in my current Vista machine, so I, unfortunately can't experience that unstable Vista that people seem to be complaining about.
The crappy state of Vista isn't due to Vista, it's due to rather poor drivers.
The exact same machine that used to hardlock once or twice a week during a long 3D gaming session on XP, has now been on 24/7 since Vista was installed without not a single hardlock.
Considering all his, you can bet Dell is chomping at the bit to get their gaming rigs switched over to Vista ASAP. Especially considering that MS will stop shipping OEM versions of Windows XP sometime this year.
Which puts AMD in a great position with regards to R600 if Nvidia isn't able to offer a stable performant driver for the G80 soon. It means that ATI has absolutely ZERO competition for space in OEM Gaming rigs.
One has to hope that Nvidia has a new more stable and robust driver coming out considering they are lauching lower end parts. If those also suffer from stability and performance issues in Vista, Nvidia stands to lose large chunks of the OEM market.
And personally I just don't think Nvidia is that stupid. So maybe there is something to the conpsiracy that Nvidia has been holding back a stable/performant driver for Vista in order to have something to make it's G8x cards look better when ATI launches R6xx series. It still seems wonky though.
Either way I expect Nvidia drivers for Vista to get significantly better soon, otherwise they are in for a world of hurt in the OEM market.
Regards,
SB
trinibwoy
13-Apr-2007, 19:45
anyone care to speculate what the R650 refresh has over the highend R600 ?
Lower power consumption seems to be the biggest win for the R650 so far. Who knows how much they could crank performance without getting back to R600 power levels. Also, it looks like the XT has 1600Mhz memory while the XTX will have 2200Mhz.
well you can exclude Dell from that OEM list of 1million + units ;) at least thats what I understand at this point.
frameavenger
13-Apr-2007, 19:49
A card slightly faster than GTX and costs less than a 640MB GTS? (And not forgetting about the IQ and other things unknown). For the first time in years I have entertained the idea of spending that much on a GPU ..
Impossible!!
My prices at launch:
ATI HD 2900XT : 580$
ATI HD 2900XTX GDDR4: 700$.
Best regards!!:grin:
trinibwoy
13-Apr-2007, 19:50
Again, please note the big IF the rumor is true that they are selling "millions" of R600s to OEMs. AMD is selling them to OEMs for a few months now, there is no doubt about that. It's the quantity that is in doubt.
Heh, considering Nvidia has moved only a couple hundred thousand G80's to this point I very much doubt R600 inventory is in the hundreds of thousand right now, far less the millions. The "millions" quote I believe came from Fudo or the Inq and that is most defintely referring to low/mid-range orders.
Heh, considering Nvidia has moved only a couple hundred thousand G80's to this point
????
think it was actually around 400k units
anyone care to speculate what the R650 refresh has over the highend R600 ? The usual tweaks, smaller process, and 50% more shaders is my guess. Orton pretty much promised the last several months ago.
flippin_waffles
13-Apr-2007, 21:35
Impossible!!
My prices at launch:
ATI HD 2900XT : 580$
ATI HD 2900XTX GDDR4: 700$.
Best regards!!:grin:
What currency? :grin:
Why is everybody so f*ing blind to talk about millions of R600s? I think INQ/FUD clearly stated that ATI was busy selling off RV610's and RV630's to OEMS.
Please stop talking about something that a 13 year old sunlight barren boy misread on a website and posted about on a forum.
Due to high OEM demands,
It looks like AMD will have to skip one more deadline. It is likely that it won't launch the RV630 and RV610 alongside the highly anticipated R600. Radeon X2900 series will officially launch in May and the NDA presentation will take place in Tunisia on the 23rd and 24th of April.
This is just one in a series of delays at ATI so we are not really surprised. RV630 and RV610 are 65 nanometres and TSMC still have some problems ramping up the production volume and yields. It is now certain that the mainstream and entry level DX 10 parts from ATI won't arrive until June, if not later.
We've also learned that Dell, HP, FSC and other big OEM's have ordered a bunch of these cards and this will also cause a delay in the retail and etail channels. We wonder what exactly ATI's partners are going to sell as ATI has delayed every single chip in the last six+ months.
http://www.fudzilla.com/index.php?option=com_content&task=view&id=499&Itemid=34
If you want to read that as AMD selling Millions upon millions of HD2900XT's and even swamping the markets in bangladesh with their $400 card, be my guest, but stop posting that nonsense and don't even let it be a base of discussion.
A brand like HP uses ATI cards exclusively in their desktop models, which indeed ship by the ten thousands a month. Something like a low profile X1300 is a cheap upgrade on those models compared to something expensive like RAM (honestly.)
The only cards nVidia sells in that market (low end cad workstations) are low end quadro's which I can imagine ATI is trying to get a foothold in, they're being outsold 3 or 4:1.
The RV630's OEMS are being used in deals for cheap multimedia/game HOME computers where brands like MEDION just like to stick 512MB of GDDR to a low core speed and sell with very low margin. these cards just have to work because they are selling them to moms and 9 year olds who want their new Vista PC to work. Stable drivers are very much a concern for a brand that doesn't want to spend a lot of money on telling people they have to update to beta drivers every other day.
Megadrive1988
13-Apr-2007, 22:20
so it seems that R650 is not really a refresh of R600, as R580 was to R520, but rather a cooler, smaller version of the R600.
I suppose R680 or R700 will be the actual refresh then.
so it seems that R650 is not really a refresh of R600, as R580 was to R520, but rather a cooler, smaller version of the R600.
I suppose R680 or R700 will be the actual refresh then.
So.. what is this R680 then? we all know that the mini core GPU's will show up within the year anyway, I hardly see a place for a (non existent) R680 in tight release schedule like that.. what about.... a R653?
Ailuros
13-Apr-2007, 23:09
I'd have a long hard laugh if we won't see in the foreseeable future a single chip ultra high end sollution from either/or IHV :P
overclocked_enthusiasm
13-Apr-2007, 23:12
I was in fact talking about not only the RV610 and RV630 but also the R600 specifically. OEMs have had their hands on R600 since the beginning of the year. OEMs according to this rumor have been ordering and receiving R600 since at least Feb./March if not earlier.
In particular OEMs have been ordering and receiving HD 2900 XTX. Notice XTX not "just" the XT that will launch first.
Which means AMD is CURRENTLY and has been for a few months making money on R600.
In other words, if the rumors are true that "millions" of R600s are being sold to OEMs that would mean that AMD has lost virtually no money by delaying R600 by approximately 2 (that is TWO) months from Feb./March timeframe to April/May timeframe.
Again, please note the big IF the rumor is true that they are selling "millions" of R600s to OEMs. AMD is selling them to OEMs for a few months now, there is no doubt about that. It's the quantity that is in doubt.
Regards,
SB
The recent warning on a revenue shortfall of over $420 million in Q1 2007 (Jan. - Mar.) simply do NOT bear your arguement out. That is a revenue shortfall of 25.45% from AMD's guidance of $1.65 billion to the recently reported $1.23 billion. Last quarter (Q4 2006 Oct. - Dec.) ATI contributed about $400 million in revenues to AMD's total of $1.77 billion.
Therefore, if you assume that ATI again contributed only $400 million to AMD's Q1 that would mean that AMD CPU sales dropped from $1.37 billion in Q4 2006 to $830 million in Q1 2007. These numbers are assuming that NO R600 sales to OEMs occured in Q1 2007.
If you really think that AMD shipped "millions" of R600 based parts to OEM's in Q1 than you are making the following case:
1. Assume ATI revenues jumped from $400 million to $600 million (estimate had to pick a number)
2. This means that of the reported $1.23 for Q1 2007, ATI is accounting for 48.8% of TOTAL revenue. This is up from 22.6% in Q4 2006.
3. This leaves only $630 million for AMD CPU sales which is a SEQUENTIAL drop of over 50%. Actually it is a 54% drop from $1.37 billion to $630 million.
I just don't see that happening. If I had to guess, and I will, ATI revenue dropped from $400 million in Q4 2006 to $375 million in Q1 2007. That means I don't think they shipped ANY R600's to OEMs and I especially don't think they shipped any 2900 XTX as this card will be held in reserve IMHO. If any R600 cards did ship in late March it was the cheaper stuff and simply replace the older RV5xx cards.
Twinkie
13-Apr-2007, 23:13
http://www.bilgiustam.com/?p=125#more-125
A benchmark featuring 8800GTX and HD 2900XT.
Test System:
Intel Core 2 Extreme QX6800
Asus P5W DH
2×1GB PC2-6400 (5-5-5-12)
Western Digital Raptor 150
Crysis demo high detail, 16x12 4xAAA 16xAF
HD2900XT - 52.8 fps
8800GTX - 50.2 fps
3dmark06
HD2900XT - 10356
8800GTX - 10951
Take it with a grain of salt though.
bigtabs
13-Apr-2007, 23:18
Ah yes the Crysis demo... :lol: I knew there was something I'd been meaning to download.
http://www.bilgiustam.com/?p=125#more-125
A benchmark featuring 8800GTX and HD 2900XT.
Test System:
Intel Core 2 Extreme QX6800
Asus P5W DH
2×1GB PC2-6400 (5-5-5-12)
Western Digital Raptor 150
Crysis demo high detail, 16x12 4xAAA 16xAF
HD2900XT - 52.8 fps
8800GTX - 50.2 fps
3dmark06
HD2900XT - 10356
8800GTX - 10951
Take it with a grain of salt though.
This is clearly a fake.
Anarchist4000
14-Apr-2007, 00:22
Maybe the numbers weren't the amount of chips that had been shipped but the amount ordered. Assuming they aren't actually getting paid until the cards get launched the impact on Q1 profits would be rather minimal. So the differences in estimated earnings could be the sales from R6xx chips being pushed into the second quarter. That would seem to line up with the differences in the forecasts as well as the millions of chips shipped/sold rumor. Also keep in mind AMD has much higher margins on CPUs(normally) and chipsets so pricing R600 low to sell more of the others in bundles isn't necessarily a bad strategy. And if it's a strong product building mindshare wouldn't hurt.
I'm also curious if the AA options on R600 are 2x, 4x, 6x, etc or 4x, 8x, 16x, and 24x. Seems like they could be quadrupling the rate they're performing AA which would make sense with the bandwidth they have. Also follows the trend of "free" 4xAA on Xenos. It's very possible the Xenos architecture was built and simply likes running 4xAA.
As for the Vista issues it's obvious OEMs have been holding back due to driver issues from Nvidia. I'm sure Nvidia alone is a major part of what has hampered the Vista launch. The stability and features of Vista are definitely better than XP but drivers are holding back the transition. I've only had one lockup due to the 7.3 BSOD issue and that's the only time my system has had any serious stability problems. Keep in mind I deal a lot with game development so I see my fair share of really bad bugs cropping up all the time. Usually it just means the display driver gets reloaded and I end up looking at my desktop.
Ailuros
14-Apr-2007, 00:54
I'm also curious if the AA options on R600 are 2x, 4x, 6x, etc or 4x, 8x, 16x, and 24x. Seems like they could be quadrupling the rate they're performing AA which would make sense with the bandwidth they have. Also follows the trend of "free" 4xAA on Xenos. It's very possible the Xenos architecture was built and simply likes running 4xAA.
Up to R5x0 Radeons were capable of 2xMSAA per cycle, meaning they needed 3 loops to reach 6xMSAA.
Now if I assume that R600 is capable of 4xMSAA per cycle it would mean that it would need 6 (!) loops in order to reach 24xMSAA.
Frankly if you chime in the necessary tiling in order to fit 4xMSAA into 720p on Xenos, I don't think it's all that "free" after all, but that's not even the point here. Xenos is a fundamentally different console design with a daughter die that contains the ROPs and (allow me to call it ) a "MSAA-framebuffer".
There's no eDRAM apparently in R600 and single cycle 4xMSAA ROPs seem to be a general trend for D3D10 GPUs if I look at G80. Question is how many loops R600 is really capable of. In any case 6 loops sounds insane for my humble layman's imagination.
Six 4-sample loops is really insane -- just imagine what EER grid will do the job for such a dense AA pattern! :shock:
I dare to dream for RGSS + MSAA combo (4xSS + 6xMS = 24), but too good to be true. :oops:
Or just throw away multisampling and leave the RGSS. :lol:
Chris123234
14-Apr-2007, 01:44
http://dailytech.com/ATI+Releases+More+R600+Details/article6903.htm
Has this info in this been discussed yet? If so point me to the location!
http://dailytech.com/ATI+Releases+More+R600+Details/article6903.htm
Has this info in this been discussed yet? If so point me to the location!
http://forum.beyond3d.com/showpost.php?p=968795&postcount=1995
:wink:
Ailuros
14-Apr-2007, 02:01
Six 4-sample loops is really insane -- just imagine what EER grid will do the job for such a dense AA pattern! :shock:
I dare to dream for RGSS + MSAA combo (4xSS + 6xMS = 24), but too good to be true. :oops:
Or just throw away multisampling and leave the RGSS. :lol:
Ironically I tried to play today NFS:Carbon in 1280*1024 with 16xS (4xRGMS + 4xOGSS). Not only did the overall result in 2048*1536 with 16xCSAA+4xTSAA look overall better, but it ended up being a lot faster too ( I wouldn't call the first case "playable" and racing sims don't demand any high framerates either).
ChrisRay
14-Apr-2007, 07:40
Ironically I tried to play today NFS:Carbon in 1280*1024 with 16xS (4xRGMS + 4xOGSS). Not only did the overall result in 2048*1536 with 16xCSAA+4xTSAA look overall better, but it ended up being a lot faster too ( I wouldn't call the first case "playable" and racing sims don't demand any high framerates either).
You know this isnt an artificial limitation right? Different GPUS with different amount of memory can go higher resolutions. Grestorm also mentioned that this limitation was inherant to 256 cards. 512/768 meg cards have different limitations for how high the XS modes can go.
Chris
Ironically I tried to play today NFS:Carbon in 1280*1024 with 16xS (4xRGMS + 4xOGSS). Not only did the overall result in 2048*1536 with 16xCSAA+4xTSAA look overall better, but it ended up being a lot faster too ( I wouldn't call the first case "playable" and racing sims don't demand any high framerates either).
Well, the 16xS pattern is quite orderish (mean - a lot more redundancy sampling), compared to the new MS modes in G80.
Anyone care to speculate when can I have R600 in my hands to test? "exact date"
Is it going to be before G80-Ultra release time frame?
Rangers
14-Apr-2007, 08:30
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
I'm still convinced R600 is a vector architecture, for what it's worth. The implication is R600 is 64 vec5. A 4:1 ALU:TEX ratio then implies 16 TMUs. And, well, 16 ROPs seems likely too.
16 TMUs fits into the ALU:TEX ratio, but it seems awfully low.
Jawed
16 TMU's is absurdly low, considering G80 has what, 32+? and these ATI "ratios" are a joke, that hopefully is good and dead.
16 TMU's would essentially mean it was little faster than R580, since that would be the bottleneck. They'd get a little bump for clock speed and maybe a little bump in shader limited titles, that's it.
Maybe, but you never know how much texture samplers are behind those 16 address units (if they are so) -- free trilinear, single cycle FP16 and so on. We'll see.
Maybe, but you never know how much texture samplers are behind those 16 address units (if they are so) -- free trilinear, single cycle FP16 and so on. We'll see.
FREE is good, but would you rather have a free DVD player in the aforementioned ferrari or are you happy with building it in a 2007 ford focus? Looking at that Dailytech article R600 is more like a funnel which is loaded with gallons of bandwidth at the top but it's being held down doing something.. free.
hence my unbalanced comment so long ago :wink:
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
I'm still convinced R600 is a vector architecture, for what it's worth. The implication is R600 is 64 vec5. A 4:1 ALU:TEX ratio then implies 16 TMUs. And, well, 16 ROPs seems likely too.
16 TMUs fits into the ALU:TEX ratio, but it seems awfully low.
With those specs, I'd expect them to get killed on the 300fps vs 200fps types of benchies. It would be interesting to see how many people fell for that as being important. If those numbers are right, they're going to need pretty demanding kind of scenarios for their advantages to show up. But then that's why one buys a high-end card, isn't it?
hence my unbalanced comment so long ago :wink:
We might be in for an interesting discussion. Let's try a thought experiment tho, just for grins and giggles.
Two applications, A & B. Two viddy cards, 1 & 2.
App A, Card 1 -- 150fps
App A, Card 2 -- 120fps
App B, Card 1 -- 50fps
App B, Card 2 -- 62fps
Now, tell me. . . which card is more "balanced"? Because my answer would be Card 2, since it provides a more consistent experience (i.e. a smaller spread between min and max) than Card 1.
Btw, I think I'm actually looking forward to [H] histograms this time quite eagerly. That's one of the things that unified is supposed to level out a bit, so it'll be interesting to see R600's vs G80's on demanding apps.
trinibwoy
14-Apr-2007, 14:18
But then you're assuming that Card 2 will only lose when there's performance to burn. i.e, your definition of a demanding app is one which taxes Card 2's strengths.
Now, tell me. . . which card is more "balanced"? Because my answer would be Card 2, since it provides a more consistent experience (i.e. a smaller spread between min and max) than Card 1.
Low median framerate, by my mean, is the most "vivid" point for smooth motion experience.
It's a bit like the V'sync + triple buffering case.
icecold1983
14-Apr-2007, 14:40
We might be in for an interesting discussion. Let's try a thought experiment tho, just for grins and giggles.
Two applications, A & B. Two viddy cards, 1 & 2.
App A, Card 1 -- 150fps
App A, Card 2 -- 120fps
App B, Card 1 -- 50fps
App B, Card 2 -- 62fps
Now, tell me. . . which card is more "balanced"? Because my answer would be Card 2, since it provides a more consistent experience (i.e. a smaller spread between min and max) than Card 1.
Btw, I think I'm actually looking forward to [H] histograms this time quite eagerly. That's one of the things that unified is supposed to level out a bit, so it'll be interesting to see R600's vs G80's on demanding apps.
card 2 is the clear choice in your example
Are both apps running under the same rendering platform (D3D or openGL) or is app. B running under openGL? ;)
nicolasb
14-Apr-2007, 15:50
A typically confused post from The Inquirer:
http://www.theinquirer.net/default.aspx?article=38923
It describes all of the different products that are supposed to feature in the R(V)6xx launch, and also talks a bit about HDMI and HD video capability.
The sting is in the tail:
The R600 1GB is very interesting. Originally, we heard about this product as a GDDR-4 only, and it is supposed to launch on Computex. We heard more details, and now you need to order at least 100 cards to get it, it will be available in limited quantities only. We expect that ATi will refrain from introduction until a dual-die product from nV shows up, so that AMD can offer CrossFire version with 2GB of video memory in total, for the same price as Nvidia's 1.5GB. Then again, in the war of video memory numbers, AMD is now losing to Nvidia flat-out.
AMD compromised its own product line-up with this 512MB card being the launch one, and no amount of marketing papers and powerpointery can negate the fact that AMD is nine months late and has 256MB of memory less than a six month old flagship product from the competitor.
bigtabs
14-Apr-2007, 15:54
(re: Inq article)
9 months late?
I thought it was only 2 months late... :?:
Unless they mean 'Late to the party' in which case - 6 months.
IIRC ATI meant to release the R580+ 9 months ago, which is what they did.
We might be in for an interesting discussion. Let's try a thought experiment tho, just for grins and giggles.
Two applications, A & B. Two viddy cards, 1 & 2.
App A, Card 1 -- 150fps
App A, Card 2 -- 120fps
App B, Card 1 -- 50fps
App B, Card 2 -- 62fps
Now, tell me. . . which card is more "balanced"? Because my answer would be Card 2, since it provides a more consistent experience (i.e. a smaller spread between min and max) than Card 1.
Btw, I think I'm actually looking forward to [H] histograms this time quite eagerly. That's one of the things that unified is supposed to level out a bit, so it'll be interesting to see R600's vs G80's on demanding apps.
Good point, I was thinking more of the lines the performance advantage won't be equal to the increased bandwidth at theoretical levels :)
bigtabs
14-Apr-2007, 17:59
That doesn't make it unbalanced though, and if the theoretical card 2 is the one with all the bandwidth, that means it's a better balanced card.
depends on how you look at it the x1800xt was a great card with bandwidth usage optimizations but if you get a game that is shader and fillrate bound......
This might be the same situatation we see the r600 in but of course to a lesser degree. This is why we see a 1 to 1 increase with the g80 when you clock the core up and around .5 to 1 increase when memory is clocked up, there was an article with overclocking on the g80 where they showed this. I don't remember the exact ratios but something close to this. The g80 isn't fully bandwidth bound, so the gains we see with the extra bandwidth on the r600, if the r600 doesn't have equal fillrates and shader through put, will be diminishing returns.
edit if I remember correctly after thinking about it, it was an article that showed both bandwidth and fillrate due to over clocking gave a 1 to 1 increase on the g80.
bigtabs
14-Apr-2007, 19:07
depends on how you look at it the x1800xt was a great card with bandwidth usage optimizations but if you get a game that is shader and fillrate bound......
I was of the understanding that the X1800XT was performing very nicely in modern titles against the 7900GTX.
... we see a 1 to 1 increase with the g80 when you clock the core up and around .5 to 1 increase when memory is clocked up....
The g80 isn't fully bandwidth bound, so the gains we see with the extra bandwidth on the r600, if the r600 doesn't have equal fillrates and shader through put, will be diminishing returns.
So the parts where the G80 is bandwidth bound are to be removed by the massive bandwidth advantage of the R600. 0.5 - 1 to 1 you say? Hardly diminishing is it? Though it is lower than 1 to 1 as you say.
I'm not sure you meant diminishing returns, as much as you meant, or at least meant to imply ;) low return. Even so... it's still quite a hefty return by your calculations wouldn't you say?
Rangers
14-Apr-2007, 19:17
A typically confused post from The Inquirer:
http://www.theinquirer.net/default.aspx?article=38923
It describes all of the different products that are supposed to feature in the R(V)6xx launch, and also talks a bit about HDMI and HD video capability.
The sting is in the tail:
512 MB seems to be kind of the sweet spot, though.
I dont think you really need more currently+near future.
So it might actually kind of be a good move, if it allows them to cut costs.
Robin B
14-Apr-2007, 19:19
So all the bandwidth the R600 has....is for no good? Why on earth do it have a 512 bit bus then ??? It seams to me it is just to waste...
PatrickL
14-Apr-2007, 19:30
Keep in mind it is just a rumour thread :wink:
With HDR becoming more and more common, it's likely that the biggest use of the bandwidth will come from having 64-bit+AA displays, and/or deferred rendering with AA.
bigtabs
14-Apr-2007, 19:35
'Deferred rendering with AA'. - I let out a contented sigh whilst reading that.
Something to look forward to with DX10 for sure.
I was of the understanding that the X1800XT was performing very nicely in modern titles against the 7900GTX.
So the parts where the G80 is bandwidth bound are to be removed by the massive bandwidth advantage of the R600. 0.5 - 1 to 1 you say? Hardly diminishing is it? Though it is lower than 1 to 1 as you say.
I'm not sure you meant diminishing returns, as much as you meant, or at least meant to imply ;) low return. Even so... it's still quite a hefty return by your calculations wouldn't you say?
Well from a theoretically stand point the r600 should have at least a 50% advantage if it going with 140+ bandwidth actually should be more like 75% advantage, but what if it ends up at 20% only in real world testing? Thats why I say diminishing returns.
Silent_Buddha
14-Apr-2007, 23:22
Well from a theoretically stand point the r600 should have at least a 50% advantage if it going with 140+ bandwidth actually should be more like 75% advantage, but what if it ends up at 20% only in real world testing? Thats why I say diminishing returns.
However, what if the advantage is 20% in real world applications when both cards are doing 8x AA and 16x AF at 1920x1200 res.
And what if the advantage remains 20% if the ATI card is doing 16x AA and 16x AF while the competition is doing 8x AA and 16x AF at 1920x1200 res?
And what if the advantage remains 20% if ATI is doing 24x AA/16x AF while Nvidia is doing 16X AA/16X AF at 1920x1200 res?
And what if the advantage is 0% for ATI if there is no AA and no AF at 1280x1024 res?
Or how about if ATI has an greater than 20% advatage at 2560x1600 res with AA/AF?
Not to say that any of that is true, but until such time as various scenarios can be tested, it's rather difficult to come out with a blanket statement that the bandwidth will only afford XX% of an advantage.
Myself I'll just sit and wait to see how it performs in various situations before declaring it a good idea or a bad idea for them to have done this. Granted much of that bandwidth has been eroded by going with GDDR3 for the initial launch.
Regards,
SB
edit if I remember correctly after thinking about it, it was an article that showed both bandwidth and fillrate due to over clocking gave a 1 to 1 increase on the g80.I'd like to find an article that shows how G80 responds just to higher memory clocks. (IIRC, I Googled for one for a reply to Shtal, possibly earlier in this thread.) All I could find was an Anandtech article that overclocked the ALUs and the rest of the GPU separately, demonstrating that it responded better to more fillrate than to more shader power.
hmm looking myself I came across that anandtech article too, weird I remember there was at least one that was done, if I can't find it by morning tomorrow, I'll run a test with Oblivion tomorrow and do a quick analysis.
trinibwoy
15-Apr-2007, 05:52
I'm surprised the B3D guys didnt do something like that in the performance review. The AF impact graphs were cool but considering how much we've debated G80's bandwidth (in)sufficiency it's strange that no attention was paid to the issue in the article.
WE HAVE CONFIRMATION of AMD's final naming convention for the R600. And the top end card is now the ATI Radeon HD 2900 XT.
The new series of cards, including the 2600, will drop the 'X' that has accompanied the model number since the X800, and harks back to the heady days of the Radeon 9700 series. http://www.theinquirer.net/default.aspx?article=38911
(I though X was = to 10) that what it should have represented at the first place anyway....
"Original Idea!"
Radeon 7000, 7200, 7500 was represented DirectX 7
Radeon 8500 was represented DirectX 8
Radeon 9500, 9600, 9700, 9800 was represented DirectX 9
X = 10 :) So Radeon X600, X700, X800 should have represented DirectX 10, but their was no support for SM3 or SM4
Radeon X1600, X1800, X1900 should have represented DirectX 11 SM5
I think ATI finally realize that they are no longer following Microsoft DirectX path steps anymore.
And here we go again! No X. :) yahooo :lol:
Radeon 2600, 2900 series; what's next - no X anymore at the front!
But next will be 3xxx series, 4xxx series, 5xxx series, 6xxx series and then we are going to reach to Radeon 7000, 7200, 7500 series, number scheme!
Geeforcer
15-Apr-2007, 07:35
They haven't been following Direct X since Radeon 9000/9200. :huh:
They haven't been following Direct X since Radeon 9000/9200. :huh:
I know it was their first misleading number scheme, especially renamed Radeon 8500 to 9100....
That will happen in a year or so and G80 will be very old by then.
http://www.xbitlabs.com/images/video/msi8800gts-640/games/sam_4x.gif
Based comparison with Geforce 8800GTS 320MB vs. 640MB
I see how 8800GTS-320MB suffers from limitation the amount of memory, because this game requires minimal 512MB video memory.
The Croatian developer Croteam’s project demands that the graphics card had 512 megabytes of memory. If the card doesn’t meet this requirement, its performance is going to be horribly low. http://www.xbitlabs.com/articles/video/display/msi8800gts-640_13.html
I wonder what would be minimal requirements for upcoming Crysis game.
It will be very interesting R600 1GB video memory vs. G80 768MB video memory.
Silent_Buddha
15-Apr-2007, 08:31
That's one of the things that might make me hold off on HD 2900 XT. With a few games on the horizon taking advantage of up to 1 gig of memory on the video card (UE3 is a prominent one) I'm a bit hesitant. Optimally this next card will last me 2 years, but I'll be happy with 1.5 years.
Still that 400 USD price tag makes it tempting to grab it as an impulse buy rather than waiting for the XTX.
Regards,
SB
That's one of the things that might make me hold off on HD 2900 XT. With a few games on the horizon taking advantage of up to 1 gig of memory on the video card (UE3 is a prominent one) I'm a bit hesitant. Optimally this next card will last me 2 years, but I'll be happy with 1.5 years.
Still that 400 USD price tag makes it tempting to grab it as an impulse buy rather than waiting for the XTX.
Regards,
SB
I'm also interesting if their is going to be any impact for near future 384bit vs. 512bit memory bandwidth - to see how extra bandwidth will serve for ATI/AMD.
Perhaps DX10 memory managment + PCIe 2.0 will help ease the burdon on board graphics memory...
That graph is interesting. Are the lower numbers minimum frame rate? The 8800 seems to be consistently very low there... ? It really does show how well the x1950 scales though. It's not *that* much slower, and at the highest resolution and it's bang on the stock GTS (pretty amazing actually)
Perhaps DX10 memory managment + PCIe 2.0 will help ease the burdon on board graphics memory...
That graph is interesting. Are the lower numbers minimum frame rate? The 8800 seems to be consistently very low there... ? It really does show how well the x1950 scales though. It's not *that* much slower, and at the highest resolution and it's bang on the stock GTS (pretty amazing actually)
I'm not going to paste directly more examples here, but if you click on that link, it shows more examples how GTS-320MB suffers based amount memory it has.
http://www.xbitlabs.com/articles/video/display/msi8800gts-640_9.html
http://www.xbitlabs.com/articles/video/display/msi8800gts-640_11.html
http://www.xbitlabs.com/articles/video/display/msi8800gts-640_15.html
http://www.xbitlabs.com/articles/video/display/msi8800gts-640_16.html
http://www.xbitlabs.com/articles/video/display/msi8800gts-640_19.html
http://www.xbitlabs.com/articles/video/display/msi8800gts-640_21.html
But in 3Dmark it does not suffers from that issue, only it does at extreme settings.
{Edit: if you look at the graph it shows 7FPS minimum and 30FPS average. "Serious Sam 2 game chart" }
Gelanin
15-Apr-2007, 10:59
http://www.xbitlabs.com/images/video/msi8800gts-640/games/sam_4x.gif
Based comparison with Geforce 8800GTS 320MB vs. 640MB
I see how 8800GTS-320MB suffers from limitation the amount of memory, because this game requires minimal 512MB video memory.
I wonder what would be minimal requirements for upcoming Crysis game.
It will be very interesting R600 1GB video memory vs. G80 768MB video memory.
Hmmm, are the "abyssmal" G80 drivers the most likely culprit for its really low minimum numbers ?
Both the X1950XTX and the 7950GX2 seems to do alot better in the minimum framerate department, which is surprising, considering how most thought Unified shaders should smooth out fps curves.
http://we.pcinlife.com/thread-749836-1-1.html
http://img341.imageshack.us/img341/639/attachment239ce1nt9.jpg
http://we.pcinlife.com/thread-749836-1-1.html
http://img341.imageshack.us/img341/639/attachment239ce1nt9.jpg
R600 series? Hopefully it's the XL or Pro then. :)
leoneazzurro
15-Apr-2007, 11:58
It's a E6600 - default at 1280x1024.
So not to shabby even if it's a XT
R600 series? Hopefully it's the XL or Pro then. :)
driver says 8.351, (http://img341.imageshack.us/img341/639/attachment239ce1nt9.jpg) remember fudzilla's item about 8.361 providing a lot of improvements over that version...(8.351) (http://www.fudzilla.com/index.php?option=com_content&task=view&id=525&Itemid=1)
Current highest score for non-OC 6600 plus SLI 8800GTX is 11339 (http://service.futuremark.com/compare?3dm06=1556056)
I know it was their first misleading number scheme, especially renamed Radeon 8500 to 9100....
Actually at first everyone thought that the first number stood for the DX version that was common at that time. But later on ATi stated that this wasn't true and the first number stood for the generation of ATi GPU. So the 7x00 serie was based on the 7th ATi GPU ever. The 8x00 serie was based on the 8th ATi GPU generation. Etc etc etc. This is of course not taking into account any crappy renaming stuff that they pulled (eg 9600 -> X300 -> X550 -> X1050 or 8500 -> 9100).
The removal of the X and replacing it by HD looks to me like this was a decision made by AMD management. It wasn't until long ago that it still was X2900, which you tell by the fact that even the AMD website mentioned (http://uk.theinquirer.net/?article=38747) the X2900. Now it's time to let go of all we ever knew about the naming scheme. It's a brand new start. nVidia will have a GF9700 and or GF9800 at the end of this year/beginning of next year and who knows what will happen next.... GF X800 series?
According to the guy in this post (you may need a reviewer permission to see the stuff):
http://we.pcinlife.com/attachment.php?aid=731906&noupdate=yes¬humb=yes
The card is a 2900XTX 1G DDR4 version, and in Windows XP with a OCed E6600 (edited, it is a E6600@300x9 as the guy suggested), it got a 3D MARK 06 of 10572.
This is of course not taking into account any crappy renaming stuff that they pulled (eg 9600 -> X300 -> X550 -> X1050 or 8500 -> 9100).
9600 became x600
The removal of the X and replacing it by HD looks to me like this was a decision made by AMD management. It wasn't until long ago that it still was X2900, which you tell by the fact that even the AMD website mentioned (http://uk.theinquirer.net/?article=38747) the X2900.
True, they allready released the X2300, so a HD2400 is a definite derivative.
But What about SmartShader, SmoothVision, Videoshader etc. they allready had the HD moniker in the x1k series...
Anarchist4000
15-Apr-2007, 13:45
Out of curiosity wouldn't it be possible that ATI tweaked the early drivers so they didn't work well with R600 to prevent the early release of benchmarks? Sort of how they did the memory controller optimizations in the past but in reverse.
I'd imagine the HD naming has more to do with AMD attempting to sell the HD video acceleration as the main feature over speed or performance of the card. Plus it would get to damn hard to differentiate between CPUs and GPUs for the average consumer the way they've been naming things.
epicstruggle
15-Apr-2007, 14:01
I'd imagine the HD naming has more to do with AMD attempting to sell the HD video acceleration as the main feature over speed or performance of the card. Plus it would get to damn hard to differentiate between CPUs and GPUs for the average consumer the way they've been naming things.
I think thats a plausible explanation, true HD gaming has finally come of age.
driver says 8.351, (http://img341.imageshack.us/img341/639/attachment239ce1nt9.jpg) remember fudzilla's item about 8.361 providing a lot of improvements over that version...(8.351) (http://www.fudzilla.com/index.php?option=com_content&task=view&id=525&Itemid=1)
Current highest score for non-OC 6600 plus SLI 8800GTX is 11339 (http://service.futuremark.com/compare?3dm06=1556056)
1000k+ with a newer driver this close to launch is impressive. ;)
I hope they haven't pulled an NVIDIA ala' NV3x.
trinibwoy
15-Apr-2007, 14:16
Well we haven't had a driver cheating fiasco in a while and you know how much everybody loves those :lol:
Colored mipmaps ftw! But seriously, I think those scars run deep and won't soon be forgotten.
Well we haven't had a driver cheating fiasco in a while and you know how much everybody loves those :lol:
Colored mipmaps ftw! But seriously, I think those scars run deep and won't soon be forgotten.
I heard the 8.371 launch drivers will only draw wireframes and all shader programs will be replaced with color values of 0,0,0.
Only in those circumstances will the XTX be 20% faster than overclocked GTX's
Heh, considering Nvidia has moved only a couple hundred thousand G80's to this point
'Only'? :???: That's huge for the high-end card.
I'm sure Nvidia alone is a major part of what has hampered the Vista launch.
Revenge for the ATI getting the X360 deal? :twisted:
Cuthalu
15-Apr-2007, 14:39
driver says 8.351, (http://img341.imageshack.us/img341/639/attachment239ce1nt9.jpg) remember fudzilla's item about 8.361 providing a lot of improvements over that version...(8.351) (http://www.fudzilla.com/index.php?option=com_content&task=view&id=525&Itemid=1)
Current highest score for non-OC 6600 plus SLI 8800GTX is 11339 (http://service.futuremark.com/compare?3dm06=1556056)
But single 8800gtx gets ~10000, doesn't look very good for that card if it's XT.
trinibwoy
15-Apr-2007, 14:41
'Only'? :???: That's huge for the high-end card.
Yeah I know. The "only" was in context of the claimed millions of orders. According to JPR Q4 numbers only 450,000 cards in the $250-700 range were shipped in Q4.
http://www.xbitlabs.com/news/video/display/20070413235044.html
But single 8800gtx gets ~10000, doesn't look very good for that card if it's XT.
Apple's Oranges?
Fastest 2.4Ghz CPu with 8800GTX's in SLI scores 12981 points (http://service.futuremark.com/compare?3dm06=1238049)
Can you show me scores for a E6600 running a single 8800 attaining that score?
Yeah I know. The "only" was in context of the claimed millions of orders. According to JPR Q4 numbers only 450,000 cards in the $250-700 range were shipped in Q4.
http://www.xbitlabs.com/news/video/display/20070413235044.html
What millions of orders? of sales of RV610? no one ever claimed selling millions of high end cards here..
Cuthalu
15-Apr-2007, 14:59
Apple's Oranges?
Fastest 2.4Ghz CPu with 8800GTX's in SLI scores 12981 points (http://service.futuremark.com/compare?3dm06=1238049)
Can you show me scores for a E6600 running a single 8800 attaining that score?
Yes, that was with E6600 and Vista. Here's one of the many: http://service.futuremark.com/compare?3dm06=1279169
trinibwoy
15-Apr-2007, 15:09
What millions of orders? of sales of RV610? no one ever claimed selling millions of high end cards here..
I was responding to this (my bolding) :
Again, please note the big IF the rumor is true that they are selling "millions" of R600s to OEMs.
Unknown Soldier
15-Apr-2007, 15:37
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
I'm still convinced R600 is a vector architecture, for what it's worth. The implication is R600 is 64 vec5. A 4:1 ALU:TEX ratio then implies 16 TMUs. And, well, 16 ROPs seems likely too.
16 TMUs fits into the ALU:TEX ratio, but it seems awfully low.
Jawed
Hmm.. HD2900XT to have 64 shaders only while the HD2900XTX to have 128 Shaders?
US
Hmm.. HD2900XT to have 64 shaders only while the HD2900XTX to have 128 Shaders?
US
Ehm.. 64 what? 128 what? that 64 would be 64x Vec4+Scalar or 64x Vec5, as we know they have "320 MADD units"
How do you fit 128 into that?
trinibwoy
15-Apr-2007, 15:56
Hmm.. HD2900XT to have 64 shaders only while the HD2900XTX to have 128 Shaders?
US
XTX probably just has higher clocks.
Unknown Soldier
15-Apr-2007, 15:56
Ehm.. 64 what? 128 what? that 64 would be 64x Vec4+Scalar or 64x Vec5, as we know they have "320 MADD units"
How do you fit 128 into that?
Everyone expects 128 Unified Shaders but R600 doesn’t have more than 64. It seems to be enough to make Nvidia G80 GTX run for its money. Nvidia has 128 unified Shaders but we are puzzled where does the difference comes from.
The Shaders are divided in four blocks and each has 16 units.
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
US
Unknown Soldier
15-Apr-2007, 15:58
XTX probably just has higher clocks.
Ye.. I would agree except that they mentioned a while ago that all the R600's would run the same clocks. Of course that could've changed since then .. but wouldn't the 800 core XT with GDDR3 still be slower than 800 core XTX with GDDR4?
Seems to me that it would.
US
trinibwoy
15-Apr-2007, 16:14
Yeah my recollection of the rumours is that the XTX and XT would be at different clocks. Not really sure though. And nobody was expecting 128 shaders, the Inq was just being asinine as usual. They probably mean that people are expecting 128 because g80 has 128.
According to the guy in this post (you may need a reviewer permission to see the stuff):
http://we.pcinlife.com/attachment.php?aid=731906&noupdate=yes¬humb=yes
The card is a 2900XTX 1G DDR4 version, and in Windows XP with a OCed E6600 (edited, it is a E6600@300x9 as the guy suggested), it got a 3D MARK 06 of 10572.
Here's the screenshot of that score.
http://img404.imageshack.us/img404/6958/attachment13b44fsm3.jpg
1024 MB. That alone makes my hangover to feel less sucky. :wink:
Unknown Soldier
15-Apr-2007, 16:22
Yeah my recollection of the rumours is that the XTX and XT would be at different clocks. Not really sure though. And nobody was expecting 128 shaders, the Inq was just being asinine as usual. They probably mean that people are expecting 128 because g80 has 128.
Ye I suppose... I guess covering their ars.. erm .. I mean bases.
US
http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
US
Everyone is expecting... why? Because 8800 has?
That's just bullcrap and nonsense, it's irrelevant "how many units" it has compared to 8800, when they're not even the same type of units (scalar vs vec4+scalar or vec5)
vertex_shader
15-Apr-2007, 16:32
I heard the 8.371 launch drivers will only draw wireframes and all shader programs will be replaced with color values of 0,0,0.
Only in those circumstances will the XTX be 20% faster than overclocked GTX's
No, AMD not cheating with any r600 drivers.
XT already faster 20% in some circumstances than the 8800 gtx oc.
vertex_shader
15-Apr-2007, 16:33
Here's the screenshot of that score.
http://img404.imageshack.us/img404/6958/attachment13b44fsm3.jpg
1024 MB. That alone makes my hangover to feel less sucky. :wink:
Funny :wink:
INKster
15-Apr-2007, 17:00
No, AMD not cheating with any r600 drivers.
XT already faster 20% in some circumstances than the 8800 gtx oc.
We won't know that for sure until the tests prove consistency.
I was responding to this (my bolding) :
Sorry, I allready posted it a couple of pages down, someone misquoted fudzilla when it reported that the reason of the launch delay is because ati is busy with the OEM orders for RV610's, which (logically) is normal business. No one ever (even at the level of Inq) ever said amd sold or had pre-orders for millions of R600's
Yes, that was with E6600 and Vista. Here's one of the many: http://service.futuremark.com/compare?3dm06=1279169
Ah sorry.. I had my search on FM approved drivers ;)
Even though it says stock, there is still a lot of variance in memory speed on those gtx's
http://www.bjorn3d.com/read_pf.php?cID=1062
right around 10k for stock 8800 gtx on a 6600e
leoneazzurro
15-Apr-2007, 18:14
http://www.bjorn3d.com/read_pf.php?cID=1062
right around 10k for stock 8800 gtx on a 6600e
That score is at 1024x768, the other score is at 1280x1024
INKster
15-Apr-2007, 18:31
That score is at 1024x768, the other score is at 1280x1024
No, it isn't.
"Default settings" means 1280x1024 in 3DMark 2006.
leoneazzurro
15-Apr-2007, 18:59
No, it isn't.
"Default settings" means 1280x1024 in 3DMark 2006.
Sorry then, I apologize.
icecold1983
15-Apr-2007, 20:28
i can confirm 10000 with a stock 8800 + e6600. thats my exact setup and i scored 9995 in 3dmark 06 at default.
Hmmm, are the "abyssmal" G80 drivers the most likely culprit for its really low minimum numbers ?
Both the X1950XTX and the 7950GX2 seems to do alot better in the minimum framerate department, which is surprising, considering how most thought Unified shaders should smooth out fps curves.
I truly understand your point and you are correct; but you missing whole point, what I was trying to compare that some games @ extreme settings needs 512MB+ video memory and 8800GTS-320MB suffers based on that requirements. But behind my whole point was that R600 1GB memory may have a huge advantage over 8800GTX-768MB memory, and some time near future when @ extreme setting minimum requirements be 1GB video memory for games - maybe even Crysis will push 8800GTX to its limits. So to make a bit clear that R600 will last a lot longer before it becomes obsolete.
INKster
15-Apr-2007, 22:33
I truly understand your point and you are correct; but you missing whole point, what I was trying to compare that some games @ extreme settings needs 512MB+ video memory and 8800GTS-320MB suffers based on that requirements. But behind my whole point was that R600 1GB memory may have a huge advantage over 8800GTX-768MB memory, and some time near future when minimum requirements be 1GB video memory for games - maybe even Crysis will push 8800GTX to its limits. So to make a bit clear that R600 will last a lot longer before it becomes obsolete.
1GB video RAM as a minimum requirement ?
Yeah, good luck with that.
By the time it happens, DX10 will be little more than a distant memory.
1GB video RAM as a minimum requirement ?
Yeah, good luck with that.
By the time it happens, DX10 will be little more than a distant memory.
http://www.xbitlabs.com/images/video/msi8800gts-640/games/6dm3_4x.gif
I was talking minimum based @ the setting you are, but not as minimum all the time.
Just like example in 3DMark06 testing 8800GTS-320MB
1GB video RAM as a minimum requirement ?
Yeah, good luck with that.
By the time it happens, DX10 will be little more than a distant memory.
That's what I meant too, I guess he's talking about the case of buying the card and using it for 1.5-2 years or so, in which case it would somewhat matter to the "average" users. The hardcore bunch will have the next big thing by then anyway, so I fail to see any relevance.
That's what I meant too, I guess he's talking about the case of buying the card and using it for 1.5-2 years or so,
Correct! :) That's why I still have AIW Radeon 9800Pro. Plus I bought AIW Radeon 9700PRO when it came out and I used it until it died :lol:
But I bought AIW Radeon X800XT and it was crap! to much problems with AIW features/software/drivers.
INKster
15-Apr-2007, 22:53
That's what I meant too, I guess he's talking about the case of buying the card and using it for 1.5-2 years or so, in which case it would somewhat matter to the "average" users. The hardcore bunch will have the next big thing by then anyway, so I fail to see any relevance.
In which case it is the raw performance of the GPU, not necessarily the amount of memory, that is responsible for such longevity.
In which case it is the raw performance of the GPU, not necessarily the amount of memory, that is responsible for such longevity.
I think you need to check your vision, how would you explain 8800GTS-640MB vs. 320MB. Same GPU!!!!!!
INKster
15-Apr-2007, 23:15
I think you need to check your vision, how would you explain 8800GTS-640MB vs. 320MB. Same GPU!!!!!!
Different price (about $100, not exactly small change), for starters ?
You have to remember that 320MB is only 64MB above the 256MB level which is the minimum required by most current games at mid-to-top quality settings.
Meanwhile, the 640MB version sometimes loses to a 512MB X1950 XTX, and it has 384MB more of memory than the current 256MB minimum level.
The answer is balance. Balance between raw power, memory bandwidth and memory capacity.
This difference would probably be explained better if we had a 384MB 8800 GTX and a few games detailing the shader vs texture view from different PoV's.
Meanwhile, the 640MB version sometimes loses to a 512MB X1950 XTX,
That is because for particular situations you don't need more then 512MB video memory, but it rely on Raw power of the GPU instead. (edit: 8800GTS vs. X1950XTX have same amount of memory bandwidth to 64GB vs. 64GB)
This is how simple the math is ? 8800GTS-640MB and 8800GTS-320MB shares same frequency and raw power of the GPU, but in some situations when texture quality/requirements changes and it needs more amount of memory; then 640MB 8800GTS version plays into advantage.
Why do we care about how cards score with uncompressed textures? I'm under the impression that there's virtually no visible difference b/w Doom 3's HQ and Ultra or SS2's High and Extreme texture settings, so really nothing to show for eating up all that memory and bandwidth but gnarlier chart titles and occasional framerate cliffs.
Isn't HDR more interesting than uncompressed textures? When devs target standard 128, 256, or 512MB RAM sizes, do they now account for AA and now HDR buffers? If not, a 640-768MB G80 may snag some advantage over a 512MB card with both of those features enabled. I'm skeptical we'll see a huge difference b/w 768MB and 1GB cards in a significant number of games until we see a lot more of them around, and I doubt a whispered 1.5GB card will do anything but pad the MSRP compared to a 1GB one for even longer.
The 320MB 8800GTS isn't a revelation b/c it's a high-end card, but without the 512MB that's been standard for high-end cards for a while, like INKster said.
"Everyone expects 128 Unified Shaders but R600 doesn’t have more than 64. It seems to be enough to make Nvidia G80 GTX run for its money. Nvidia has 128 unified Shaders but we are puzzled where does the difference comes from." --http://www.fudzilla.com/index.php?option=com_content&task=view&id=526&Itemid=1
LO...L? That one crosses from funny to sad. You can't report on this stuff for so long and be puzzled by the difference. Maybe the laugh's on me, for not realizing why pretending to not know the difference is advantageous.
Why do we care about how cards score with uncompressed textures? I'm under the impression that there's virtually no visible difference b/w Doom 3's HQ and Ultra or SS2's High and Extreme texture settings, so really nothing to show for eating up all that memory and bandwidth but gnarlier chart titles and occasional framerate cliffs.
Isn't HDR more interesting than uncompressed textures? When devs target standard 128, 256, or 512MB RAM sizes, do they now account for AA and now HDR buffers? If not, a 640-768MB G80 may snag some advantage over a 512MB card with both of those features enabled. I'm skeptical we'll see a huge difference b/w 768MB and 1GB cards in a significant number of games until we see a lot more of them around, and I doubt a whispered 1.5GB card will do anything but pad the MSRP compared to a 1GB one for even longer.
The 320MB 8800GTS isn't a revelation b/c it's a high-end card, but without the 512MB that's been standard for high-end cards for a while, like INKster said.
advantageous.[/SIZE]
I'm not trying to cause trouble here on this forum.
But like you said right now 512MB frame buffer is standard for high-end card.
Few years ago 256MB frame buffer was standard for high-end card.
But since Nvidia made 320MB, 640MB, 768MB video memory and ATI/AMD will go from 512MB to 1024MB, I wonder what would be standard near future.
And like you said ----> High and Extreme texture settings, so really nothing to show for eating up all that memory and bandwidth but gnarlier chart titles and occasional framerate cliffs.
Also depends what future holds for extreme-quality, since Crysis game will be so much more detail especially DX10-SM4 support it may very well show it's own standard for the requirement.
Best regards :)
Shtal....
trinibwoy
16-Apr-2007, 03:41
Shtal, considering that Crytek have emphasized that Crysis will be very scalable I'm not sure why you're using it as your poster child for high memory requirements. I doubt it will have some SS2 like "512MB minimum" stipulation. It's highly unlikely that it will "need" 1GB or even 768MB at its highest settings.
Shtal, :lol: I know you're not trying to cause trouble, you troublemaker. :razz: I'm just confused by sites using these maximum texture settings in testing average framerates in high-IQ situations, and I'm not sure using uncompressed textures as a guide is helpful in determining memory size requirements. I can see devs using compressed textures and HDR buffers as memory guidelines, with AA (as common as it is) and uncompressed textures remaining as niceties they won't base their system requirements on.
As for your 3DM06 chart, I think it's an exception, not a rule. The 320MB 8800GTS is a strange one, b/c it seems like a card that may not serve the high-end gamer all that well b/c I think the definition of high-end/enthusiast cards has expanded to include HDR and AA at a high framerate, not just higher screen and texture resolution. (Did NV have more G80 GPUs to sell than RAM available to buy? If they're still selling the high-end as complete cards or GPU+RAM combos, are they trying to raise their profit margins by reducing the amount of RAM? Or, if AIBs just buy the chips and maybe the PCBs, are they the ones trying to save on RAM costs? Or are either NV or the AIBs just using this as the easiest way to fill a price gap or to fight the X1950XT/X?) I'm not paying too much attention to the GTS-320 as a guide to future cards or game memory targets, not unless NV introduces a 192-bit bus as a new mainstream class standard.
I haven't followed Crysis too closely, but why would DX10/SM4 push more texture rather than shader-based "detail?" Or do you mean devs will now consider HDR's and even AA's extra memory usage in their detail and memory targets?
Me, I think devs will still set standards based on the broadest market segments available, and 320MB is an odd bird. I can see 128MB, 256MB, 512MB, and probably 768+MB (rather than 1GB) as standard targets in the near future. I'm skeptical about 1GB being a game dev target in the next two years, but maybe I'm giving the 768MB G80 too much credit (tho 384-bit is a logical progression from 256-bit, just like 24 from 16 "pipes," I'll wait til G80's refresh or replacement to see if NV sticks with it or if it was an acceptable compromise to get G80 out on 90nm and so far ahead of R600), or Tim Sweeney's comment too little.
And remember, I approach things like a cynical skeptic, so I'm waiting for benchmarks of 1GB cards before I consider it a RAM size as significant as "768+MB."
Man, I really with 3DEclipse was still up so I could crib off of--er, refer to Ail's VRAM usage article.
Edit: Yeah, I'm using SS2 as Serious Sam 2, and I'm not sure that its uncompressed texture setting was called Extreme, not am I bothered to Google it. :)
Shtal, considering that Crytek have emphasized that Crysis will be very scalable I'm not sure why you're using it as your poster child for high memory requirements. I doubt it will have some SS2 like "512MB minimum" stipulation. It's highly unlikely that it will "need" 1GB or even 768MB at its highest settings.
Forgive me but, SS2? Are we talking about Serious Sam 2 here? I don't recall seeing a setting containing the word 'extreme' anywhere in it (as Pete mentioned). Or was the word extreme merely used to signify it's overall 'extreme' vmem usage?
Just wanted to clear that up :smile:
Shtal, considering that Crytek have emphasized that Crysis will be very scalable I'm not sure why you're using it as your poster child for high memory requirements. I doubt it will have some SS2 like "512MB minimum" stipulation. It's highly unlikely that it will "need" 1GB or even 768MB at its highest settings.
So its pretty much we have overkill video cards for at least 2 years or so!
I understand most company's make backward comparability's even to DX7 hardware, just like Half-Life2 when it came out.
I also understand that you could lower the settings and run Crysis even maybe with using Radeon 9700/9800 series, but it doesn't mean that Crysis game would not be able to stress latest hardware that is currently available. :) :)
No worry s - I'm not smart or wise!
Man, I really with 3DEclipse was still up so I could crib off of--er, refer to Ail's VRAM usage article.
:)
http://users.otenet.gr/~ailuros/Introduction.pdf
http://users.otenet.gr/~ailuros/DBPerf.pdf
http://users.otenet.gr/~ailuros/AFqnitpicks.pdf
http://users.otenet.gr/~ailuros/MCwithAA.pdf
http://users.otenet.gr/~ailuros/AAPerf.pdf
http://users.otenet.gr/~ailuros/TranspAAPerf.pdf
http://users.otenet.gr/~ailuros/Conclusion.pdf
here ya go I think these are what you are talking about, not sure though.
Dear Pete, this is my final post for this topic, but could you explain why in this game 8800GTS-320MB slower then 640MB version of GTS. Thanks....
http://www.xbitlabs.com/images/video/msi8800gts-640/games/gothic_af.gif
Despite the lack of FSAA, the GeForce 8800 GTS 320MB is far slower not only than the ordinary GeForce 8800 GTS http://www.xbitlabs.com/articles/video/display/msi8800gts-640_15.html
I haven't followed Crysis too closely, but why would DX10/SM4 push more texture rather than shader-based "detail?" Or do you mean devs will now consider HDR's and even AA's extra memory usage in their detail and memory targets?
I cannot answer this question for you right now!
As for your 3DM06 chart, I think it's an exception, not a rule. The 320MB 8800GTS is a strange one, b/c it seems like a card that may not serve the high-end gamer
Some website said can't remember which one, that if you have Dell 24" inch or especially Dell 30" inch monitor, the 320MB-version of 8800GTS will be horribly slow.
I still believe 512MB or 640MB will be next soon. "But 768MB or 1GB will be Ok for now"
Some website said can't remember which one, that if you have Dell 24" inch or especially Dell 30" inch monitor, the 320MB-version of 8800GTS will be horribly slow.
I still believe 512MB or 640MB will be next soon. "But 768MB or 1GB will be Ok for now"
I cannot imagine why a single mid-range performance part, regardless of memory could be slow for gaming 1080P or 2560x1600.
....
....
....
:roll:
Ailuros
16-Apr-2007, 05:33
http://users.otenet.gr/~ailuros/Introduction.pdf
http://users.otenet.gr/~ailuros/DBPerf.pdf
http://users.otenet.gr/~ailuros/AFqnitpicks.pdf
http://users.otenet.gr/~ailuros/MCwithAA.pdf
http://users.otenet.gr/~ailuros/AAPerf.pdf
http://users.otenet.gr/~ailuros/TranspAAPerf.pdf
http://users.otenet.gr/~ailuros/Conclusion.pdf
here ya go I think these are what you are talking about, not sure though.
Remind me to kick someone in the nuts for the total absence of 3declipse LOL :D
***edit: Pete the vram usage article on the 7800GTX isn't available anymore anyway. I can however pass you the word file if you're interested; just PM me for that.
Shtal, without knowing Xbit's settings or even Gothic 3's VRAM recommendations I can't say definitively, but it sure looks like lack of RAM. Of course, that isn't a surprise, given how little VRAM the GTS-320 has compared to the other cards. According to Gamespot (http://www.gamespot.com/pc/rpg/gothic3/tech_info.html), Gothic 3's "Recommended" VRAM is 256MB. The framerate snob's rule of thumb is twice the recommended system specs for smooth performance, so 320MB comes up short. HW.fr has such nice comparison charts (http://www.behardware.com/articles/644-22/nvidia-geforce-8800-gtx-8800-gts.html) when it comes to these things.
Just like it's generally silly to strap 512MB RAM on a 9600 (except in the rare cases like maybe MMOs that gobble up VRAM but may not need a fillrate monster to be playable), so it seems similarly silly to ship an 8800 with less than 512MB.
But I'm sure you knew all that, so I'm not sure why you asked. If your point is that more VRAM is generally safer and better, then I can agree. But I hesitate to use extreme cases like uncompressed textures for minimal IQ gain as proof of what future games will require, because those future games will probably involve a lot of other performance variables (more specific than just "they'll run slower" :wink: ). I'm sure Crysis and lots of upcoming will use more VRAM in general, and specifically will probably require more than 512MB with high settings.
Razor1, thanks much for the links. I meant the VRAM usage article Ail mentioned, but the G80 article was also great, so no harm in linking it. :)
Shtal, without knowing Xbit's settings or even Gothic 3's VRAM recommendations I can't say definitively, but it sure looks like lack of RAM. Of course, that isn't a surprise, given how little VRAM the GTS-320 has compared to the other cards. According to Gamespot (http://www.gamespot.com/pc/rpg/gothic3/tech_info.html), Gothic 3's "Recommended" VRAM is 256MB. The framerate snob's rule of thumb is twice the recommended system specs for smooth performance, so 320MB comes up short. HW.fr has such nice comparison charts (http://www.behardware.com/articles/644-22/nvidia-geforce-8800-gtx-8800-gts.html) when it comes to these things.
Just like it's generally silly to strap 512MB RAM on a 9600 (except in the rare cases like maybe MMOs that gobble up VRAM but may not need a fillrate monster to be playable), so it seems similarly silly to ship an 8800 with less than 512MB.
But I'm sure you knew all that, so I'm not sure why you asked. If your point is that more VRAM is generally safer and better, then I can agree. But I hesitate to use extreme cases like uncompressed textures for minimal IQ gain as proof of what future games will require, because those future games will probably involve a lot of other performance variables (more specific than just "they'll run slower" :wink: ). I'm sure Crysis and lots of upcoming will use more VRAM in general, and specifically will probably require more than 512MB with high settings.
You gave excellent answer and I appreciate that very much, thanks again!
And yes more video memory is generally safer/better! plus not only that but architecture design plays important role as well "Finally came to understanding :) :) :) :) "
But I hesitate to use extreme cases like uncompressed textures for minimal IQ gain as proof of what future games will require, because those future games will probably involve a lot of other performance variables (more specific than just "they'll run slower"
:( Ok....
But I'm sure you knew all that, so I'm not sure why you asked.
Correct! "The reason I said that because why would you need 512MB memory on Radeon 9600, it is plain stupid.... also why would someone do stupid thing and make G80 less then w/ 512MB memory" That's what bother me....
Me, I think devs will still set standards based on the broadest market segments available, and 320MB is an odd bird. I can see 128MB, 256MB, 512MB, and probably 768+MB (rather than 1GB) as standard targets in the near future. I'm skeptical about 1GB being a game dev target in the next two years, but maybe I'm giving the 768MB G80 too much credit (tho 384-bit is a logical progression from 256-bit, just like 24 from 16 "pipes," I'll wait til G80's refresh or replacement to see if NV sticks with it or if it was an acceptable compromise to get G80 out on 90nm and so far ahead of R600), or Tim Sweeney's comment too little.
I think you're giving the G80 too much credit. ;) The probability of the 384bit bus becoming a standard solution that is implemented across all product ranges is miniscule at best. Also, the 768MB will not become an important target for game dev's, because it is just one card. In all probability, nothing before it and nothing after it will use the same memory configuration.
My guess is that in a year's time the 8800 GTX owners will be browsing the tweak guides in a desperate search for the best settings in new games, that look good while not exceeding the 768MB limit.
If ATI/AMD makes R600 512MB and 1GB memory I'm Ok with that, as lone their mind is clear not to make R600 with only 256MB, it simply will be obsolete, nothing to add.
But if they make RV6xx with 256MB, who cares! it fits perfectly.
But G80-768MB vs. 1GB R600 I'm still forward to see that in action, but at the beginning you probably be disappointed because it is overkill right now for games.
DemoCoder
16-Apr-2007, 10:09
My guess is, in a years time, people will be wondering why they have to crank up resolution and AA to insane levels before RAM makes a difference on their next-gen cards.
The idea of a developer targeting 1Gb of VRAM is idiotic at this point.
"The reason I said that because why would you need 512MB memory on Radeon 9600, it is plain stupid.... also why would someone do stupid thing and make G80 less then w/ 512MB memory" That's what bother me....
9600 with 512 MB is just a marketing argument for the less knowledgable masses, while the 320MBG80 is there to hit a nice price point for people who want the G80 cheaper. No real sense there otherwise, just filling all possible pricing slots.
My guess is, in a years time, people will be wondering why they have to crank up resolution and AA to insane levels before RAM makes a difference on their next-gen cards.
The idea of a developer targeting 1Gb of VRAM is idiotic at this point.
I think it was Tim Sweeney who said in an interview that Unreal Engine 3 will require 1GB of video memory to run in full glory. Currently there's 25 games out or in work based on this engine, so it would not be preposterous to guess that many of these will make use of 1GB of video memory as well.
EDIT: Here's the link and the quote:
http://www.beyondunreal.com/content/articles/95_1.php
BU: UT2004 required 5.5 gigs of hard drive space to install. This has got to be a strain a lot of people's systems. So you're talking 2048 x 2048 texture sets, what kind of system and memory is this next game going to take?
TS: Well, we are aiming at the kind of PC that we think will be mainstream in 2006. We will also be able to scale it down. Basically DirectX 9 cards will be minimum spec, so any DirectX 9 shipping today will be capable of running our game, but probably at reduced detail. If you only have a 256 meg video card you will be running the game one step down, whereas if you have a video card with a gig of memory then you'll be able to see the game at full detail.
bigtabs
16-Apr-2007, 12:51
I think it was Tim Sweeney who said in an interview that Unreal Engine 3 will require 1GB of video memory to run in full glory.
Actually I believe he was referring to UT3 not UE3. A game based on the UE3 engine shouldn't require any more texture memory than the developers decide to use.
I don't think Roboblitz or Rainbow 6 Vegas require 1 gig VRAM.
I don`t think you should take what Tim Sweeney said at that point in time by heart. He`s been wrong/over-optimistic before, you know ;). There`s absolutely no indication whatsoever that suddenly, because ATi, our lords and saviours, will bring a 1 GB card to the market, everybody and their dog will start targeting that in any meaningful way. The industry doesn`t quite work like that.
That was by no means a definite statement, rather a vague estimation. Might as well be meant like the ultra-quality mode in Doom3, for example.
As for Sweeney's statements, his prediction of moving the gfx stuff to the CPU and GPU's disappearing alltogether seems to be coming along fine, looking at those Intel slides in the other thread ;)
bigtabs
16-Apr-2007, 13:45
I think it was Mark Rein who said fairly recently that 2x8800GTX in SLI with the fastest quad core available wouldn't be able to play this game maxed out.
Then he went on to state that it would be something you would be able to showcase your new hardware with for some time after release. This would indicate a higher texture requirement in my book.
Ah, here it is... LINK (http://www.gameinformer.com/News/Story/200701/N07.0126.1423.58507.htm)
GI: Unreal has always scaled really well, from low-end hardware all the way to the high end. Where do you think the sweet spot is? What do you think it takes for a rig to be able to put Unreal through all of its paces?
Rein: We always aim Unreal for systems that people don’t have yet. (laughs) Whether its UT or any Unreal game, so I think the sweet spot has yet to show up. Again, it’s 64-bit and a ton of RAM, like an NVIDIA dual 8800s and Core 2 Extreme Quad processor—you could certainly build a super rig, but UT3 with everything turned up all the way is still going to struggle on that kind of thing. A year from now, it’ll still be a game that is a showcase game for whatever hardware you happen to be getting then. That’s normal. That’s exactly the way we’ve done it every time from the original. The format hasn’t changed there. But you’re right, we try very hard to make sure it runs well on what the average gamer has. It’ll definitely be hard to reach the bottom this time, because with UT2004, we had a software renderer, so it could run on virtually anything. We’re hoping to have that capability at some point with this series, but I don’t know for sure or when.
Shtal, without knowing Xbit's settings or even Gothic 3's VRAM recommendations I can't say definitively, but it sure looks like lack of RAM. Of course, that isn't a surprise, given how little VRAM the GTS-320 has compared to the other cards. According to Gamespot (http://www.gamespot.com/pc/rpg/gothic3/tech_info.html), Gothic 3's "Recommended" VRAM is 256MB. The framerate snob's rule of thumb is twice the recommended system specs for smooth performance, so 320MB comes up short. HW.fr has such nice comparison charts (http://www.behardware.com/articles/644-22/nvidia-geforce-8800-gtx-8800-gts.html) when it comes to these things.
Just like it's generally silly to strap 512MB RAM on a 9600 (except in the rare cases like maybe MMOs that gobble up VRAM but may not need a fillrate monster to be playable), so it seems similarly silly to ship an 8800 with less than 512MB.
But I'm sure you knew all that, so I'm not sure why you asked. If your point is that more VRAM is generally safer and better, then I can agree. But I hesitate to use extreme cases like uncompressed textures for minimal IQ gain as proof of what future games will require, because those future games will probably involve a lot of other performance variables (more specific than just "they'll run slower" :wink: ). I'm sure Crysis and lots of upcoming will use more VRAM in general, and specifically will probably require more than 512MB with high settings.
Razor1, thanks much for the links. I meant the VRAM usage article Ail mentioned, but the G80 article was also great, so no harm in linking it. :)
I've been saying for months that this is going to be an interesting area to watch, as the dynamics of cost vs performance implications of various framebuffer sizes get worked out. It *isn't* that straightforward, as you've been trying to tell Shtal. Using 640MB vs 320MB as an example is not going to reliably tell you what happens with 768MB vs 512MB, or 1GB vs 768MB, or (possibly at some point) 512MB vs 384MB.
For instance, we specifically asked NV pre-release if they could point at specific current or near term scenarios where 768MB would be a performance win over 512MB, and they couldn't. Now, that doesn't mean none exist nor ever will. It probably means they have limited resources to go hunting for such (they read reviews as avidly as the rest of us, and for partly the same reason --to collect this kind of data).
Maybe they've found some since, dunno (this was 5 months ago, after all). No one should lose sight of the fact that some of these framebuffer sizes are not being driven by willy-waving, or even optimum performance considerations --possible memory configurations given bus width and memories available on the market plays a large role.
Then he went on to state that it would be something you would be able to showcase your new hardware with for some time after release. This would indicate a higher texture requirement in my book.
I'd rather guess you can turn the shadow samples and such stuff up to insane levels. I doubt they'll bother making 25 GiB of textures for a "high" setting that screams for 2 GiB of Video-RAM which you can enable 2008 or 2009.
The X2900 XT (512mb/256bit) is slower then 8800 gtx.
The 2900 XT/512mb does 10200 in 3D 06 stock clocks vs 8800 gtx @ 10900
both using QX6700 @ 3.6 testing.
http://www.xtremesystems.org/forums/showpost.php?p=2131444&postcount=29
256bit? Typo perhaps? This Dynasty guy is apparently one of those hardcore overclockers who often knows whats going on. :smile:
I'm dying here! Please release the cards! :lol:
This is just too much :)
If its true... On the one hand I'd be thoroughly disappointed, but on the other impressed that they were able to pull such 8800GTX-rivalling performance out of a 256bit card. And that would make the 512bit 2900XTX doubly worth waiting for, I guess ;)
256-bit GDDR3 card? Umm, that makes like zero sense. . .
256-bit GDDR3 card? Umm, that makes like zero sense. . .
But it does make one wonder how far would AMD go to cut the price tag on this card without hurting their own wallet in the process...
256-bit GDDR3 card? Umm, that makes like zero sense. . .
XL maybe? :???:
bigtabs
16-Apr-2007, 15:17
I'd rather guess you can turn the shadow samples and such stuff up to insane levels. I doubt they'll bother making 25 GiB of textures for a "high" setting that screams for 2 GiB of Video-RAM which you can enable 2008 or 2009.
Who mentioned 2GiB vram? I thought the discussion was about the usefulness of 1 gig. :???:
I'm sure your opinions can have their own validity without having to exaggerate mine to prove your point. :razz:
I think it was Mark Rein who said fairly recently that 2x8800GTX in SLI with the fastest quad core available wouldn't be able to play this game maxed out.
Then he went on to state that it would be something you would be able to showcase your new hardware with for some time after release. This would indicate a higher texture requirement in my book.
Ah, here it is... LINK (http://www.gameinformer.com/News/Story/200701/N07.0126.1423.58507.htm)
Good lord! Is that even...possible?! I dont recall the other UT games being this stressful on the hardware.
trinibwoy
16-Apr-2007, 16:08
256-bit GDDR3 card? Umm, that makes like zero sense. . .
I saw it too and although the guy has a good rep I assumed it was a typo. Has to be....
bigtabs
16-Apr-2007, 16:20
Good lord! Is that even...possible?! I dont recall the other UT games being this stressful on the hardware.
Indeed. They're certainly not lacking in shader grunt so where is the shortfall with that setup?
Memory perhaps? :smile:
Ok did the benchmarking of the core and ram overclock here ya guys go :) I ended up using FEAR, easier to benchmark ;)
http://www.mediafire.com/?0mtnyjxmnzt
but on the other impressed that they were able to pull such 8800GTX-rivalling performance out of a 256bit card. And that would make the 512bit 2900XTX doubly worth waiting for, I guess ;)
Not that i believe that the 256 bit rumours are true but which nr's are you using to get 8800 GTX rivaling performance ? 3D Mark ?
icecold1983
16-Apr-2007, 16:46
Ok did the benchmarking of the core and ram overclock here ya guys go :) I ended up using FEAR, easier to benchmark ;)
http://www.mediafire.com/?cylcmyvdmnt
have i entered the twilight zone?
Evildeus
16-Apr-2007, 16:47
Ok did the benchmarking of the core and ram overclock here ya guys go :) I ended up using FEAR, easier to benchmark ;)
http://www.mediafire.com/?cylcmyvdmntWhere is your R600 bench? ;)
vBulletin® v3.8.4, Copyright ©2000-2010, Jelsoft Enterprises Ltd.