View Full Version : The ATI R600 Rumours & Speculation Centrum
Pages :
[
1]
2
3
4
5
6
7
8
9
10
This thread is the new "official" place to discuss ATI's upcoming R600 chip, plus all related products and technologies, at least for the time being. The following points should be kept in mind:
The thread will be moderated to improve the signal-to-noise ratio slightly All off-topic discussions will be moved to other threads, or deleted, at our discretion This first post will be edited over time to link to related sources and leaks Further "summaries" may be based on concensus or apparent credibility, with minority reports generally noted
No matter how many stupid rules I come up with, the goal still is to have a good and active discussion going, so get started guys!
Summary
D3D10, 500M+ transistors, release sometime between January and the end of Q1 2007, certainly 80nm. Based on GDDR4, vast numbers of reports claiming a 512-bit external bus, although only midly reliable. Minority reports for 256-bit bus (although sometimes from more reliable sources), or a dual-chip solution (256-bit bus per chip, 2 chips per board).
The vast majority believe in the existence 64 unified "real pipelines" (4x16 ala Xenos' 3x16, although perhaps with different units internally). There are minority reports for 80 and 96. It is unknown what the exact organization of the pipelines will be. Xenos is Vec4+Scalar MADD per "pipeline". There are minority reports of a similar organization, as well as those of a more R580-like one, and finally also some reports or fully scalar units (4xScalar MADD + ???). None of them seem particularly more credible than the others at this point.
As per Direct3D10's requirements, native FP16 filtering should be supported, which should make Farcry HDR function on R6xx even with the original NVIDIA patch, at least in theory. Furthermore, little information is known on the AA and AF algorithms, with most believing they will be mostly similar to R580's. Minority reports of 8x MSAA support exist, however.
Reliable performance predictions are currently inexistant at this point, as only sensationalistic tabloid sites have been claiming any relative or absolute figure. It is likely that the chip will be much faster in Vertex Shading than R580, as per its unified nature. It is however unknown how high the triangle setup figures will be, as well as some other related ones such as attribute fetch. These might or might not become new bottlenecks in modern industry benchmarks. It is also unknown whether the number of TMUs and ROPs has increased from the R580's 16 each.
The power and heat figures of the R600 have been initially reported as being roughly "150W+", but recent minority reports have increased that number to 200-250W. Other older minority reports also mentionned 100-150W instead. Nothing reliable is known regarding heat dissipation and related technologies to minimize the problem, should these figures be accurate.
Links
The old R600 thread: http://www.beyond3d.com/forum/showthread.php?t=31049
incurable
16-Oct-2006, 13:58
I'm wondering about the cooling solution used, should R600 really go up into the 200W+ range, given the limitations on space available and the fact that -in a much less restrictive environment- even cooling a 130W Prescott was considered far from trivial.
Any word on a possible watercooling option on these parts (beyond the Sapphire Toxic special SKUs), as some have suggested available for G80?
Also, beyond R600 itself, I was wondering if there was anything known about its smaller brethren.
At one time it was being speculated that unification would be such a transistor-saver for low and low-mid that we might see a whole family if not simultaneously, then something close to it.
But I don't think we've heard anything about the rest of the family. The roadmap that mentions R600 in Q1 does not mention any others. . . but is labeled "High-end CrossFire Roadmap", so you wouldn't expect it to. . .
Well, as I theorised in the other thread, shouldn't we be expecting 65nm value/mainstream/performance R6xx parts during the same quarter as R600?
Additionally, if they're all (or at least some) scheduled for near-simultaneous release, would that explain the relative tardiness? Is ATI simply taking advantage of the "lateness" of Vista?
Jawed
epicstruggle
16-Oct-2006, 15:22
D3D10
Stupid question time:
Any word whether ATI plans on releasing features beyond D3D10?
trumphsiao
16-Oct-2006, 16:47
Sorry I have to state few points I suspision
1. Raw Performance of R600 or G80 would not have more than 1.5 times that of G71
2. Why It's harder to polish Winxp driver for DX10 Hardware ?
1. Raw Performance of R600 or G80 would not have more than 1.5 times that of G71So now you're telling us the G80 3DMark05/06 scores that leaked were shit? Decide yourself! ;)
2. Why It's harder to polish Winxp driver for DX10 Hardware ?It isn't. Drivers of truly new architectures always take more time to get right, that's all. Do you think the R300 and NV40 drivers were perfect overnight? Hint: They weren't. And there hasn't been such a massive architecture transition since.
Uttar
Uttar - maybe the scores were for g80 SLIed?
Uttar - maybe the scores were for g80 SLIed?
Did the news report a power surge in Taiwan ? ;)
pjbliverpool
16-Oct-2006, 23:46
I still say 64 Xenos style shaders isn't enough as it would have to run at insane clock speeds just to match R580's raw shader performance.
Plus didn't someone at ATI comment that the next gen would be over half a TFLOP? The chip would have to run at 875Mhz to achieve half a TFLOP :???:
Im placing my money on either 96 Xenos sytle shaders at 600+ Mhz or 64, significantly beefed up shaders, again at 600+ Mhz.
In fact R600 could may well just be double Xenos. 96 ALU's, 32 TMU's, 32 vertex texture units? Then either 24 or 32 ROPS and hopefully that really juicy 512bit memory interface. All running at 600+ Mhz should make quite a nice next gen chip.
Skrying
17-Oct-2006, 00:04
Does the R600 really need a 512bit external bus? I would think it'd have to be mighty powerful for all of that bandwidth.
Hence the circular nature of the discussion. Pick the rumor you believe and work it out to its natural conclusion and you get to different places.
64 units would not seem to require 512-bit gddr4. So one or the other is likely incorrect, unless someone can point at a different way to circle the square. HDR? The R580+gddr4 results would not seem to support that idea. Really eye-popping core speed? 800MHz or higher? Well, NV seems to have done something with G80 shaders to get to 1350MHz, so maybe that possibility shouldn't be entirely dismissed, tho the conventional wisdom doesn't seem to favor it.
Oh yeah, toss some fairly frightening heat/power rumors in there too in trying to adjudicate the rumor mill. . . but that could go either way, either > 64 shaders or 800MHz+ clock. Would the power/heat rumors be consistent with 64 shaders at 650MHz? I tend to think not.
I really can't understand why 64 pipes has any credibility.
If Orton winked when he said 96 pipes recently, why are people still beating round the bush?
Jawed
Oh, hang on. Orton winked and said 96? When/where was this?
The 64 comes from xbit claiming that ATI tipped them that way.
http://www.techreport.com/etc/2006q4/stream-computing/index.x?pg=1
Jawed
Son of a gun.
Orton pegged the floating-point power of today's top Radeon GPUs with 48 pixel shader processors at about 375 gigaflops, with 64 GB/s of memory bandwidth. The next generation, he said, could potentially have 96 shader processors and will exceed half a teraflop of computing power.
Missed that. Right. Time perhaps to update the conventional wisdom watch then, even with that "potentially" hedge.
It's why 512-bit has credibility for me, even though I'm dubious about die size being big enough for such a big bus, particularly with 65nm ~1 year away.
Jawed
PeterAce
17-Oct-2006, 00:31
So R500/C1 is 500 * (48 * 9) = 216000
So by that logic R600 might be:
600 * (96 * 9) = 518400 ;)
So 600 Mhz and six 16-way shader arrays. Very nice if true.
They can always shrink back to a smaller bus size and over-compensate with insane memory clocks, no?
Skrying
17-Oct-2006, 00:38
They can always shrink back to a smaller bus size and over-compensate with insane memory clocks, no?
I wonder how that'd go over. Use a massive bus till you get insane memory clocks. Interesting, but I personally dont think it'd make sense. The marketing would be terrible, the layout for 512bit would be insane......... wait, does sorta sound like ATi.
Maybe, but your PR people will hate you. They hate getting everyone all ZOMG about something and then explaining later why the next part is still ZOMG when they take it away.
Not impossible, but generally not a happy prospect. And gddr4 is the big bump, so I dunno where you'd suddenly get that extra bw post gddr4. . .
Ailuros
17-Oct-2006, 00:44
They can always shrink back to a smaller bus size and over-compensate with insane memory clocks, no?
Sounds like a huge waste of R&D resources if I didn't misunderstand you. I'm quite confident that IHVs know the limits for each future architecture right from the start.
It's why 512-bit has credibility for me, even though I'm dubious about die size being big enough for such a big bus, particularly with 65nm ~1 year away.
The harder I look at what TR reported Orton saying, the more I notice that he gave current bw, current flops, future flops, and. . .oh, wait, no future bw. But if you extrapolate from what he did give, you'd be lead to believe a number in the 120GB/s range. It's arranged in classic "what's the missing number" fashion almost.
Ha.
Hmm I was thinking something like xdr2. There's also some talk of GDDR5 already so who knows (maybe that'll reduce the pin requirements versus GDDR4?). I guess it's more likely for the ultra high-end chips to just keep pushing the die-size envelope ...
Bouncing Zabaglione Bros.
17-Oct-2006, 01:03
Oh, hang on. Orton winked and said 96? When/where was this?
The 64 comes from xbit claiming that ATI tipped them that way.
If you were ATI and wanted to blindside an earlier launching G80 by leaking a low number that would have Nvidia thinking they'd got you cold, only to come back with R300 style insanity....?
Well, I'm careful to describe stuff as "conventional wisdom", not "received truth". :lol: Anyone who lived thru "extreme pipes" is not going to forget that experience. :cool:
PeterAce
17-Oct-2006, 01:11
So R500/C1 is 500 * (48 * 9) = 216000
So by that logic R600 might be:
600 * (96 * 9) = 518400 ;)
So 600 Mhz and six 16-way shader arrays. Very nice if true.
And wrt the memory controller I hope that it is a 384-bit external memory bus being - 12 memory channels (6 primary-ring stops and two channels per ring stop).
Which fits nicely with a primary-ring stop per 'shader processor' (each one being 4 TMUs, 16 ALUs, 4 ROPs). In other words an primary-ring stop per 'heavy client'.
Definitely worth holding onto as a possible configuration.
Jawed
Ailuros
17-Oct-2006, 01:26
Frankly gentlemen I couldn't give a rat's behind, if a GPU has a 256-bit bus with 1.4GHz GDDR4 or a 384-bit or even 512-bit bus with whatever GDDR3, as long as it's delivering the performance I'd expect. Last case scenario sounds costly and complicated, but if the total price/performance ratio matches in the end I don't see a problem there nor would I judge a GPU by it's buswidth.
INKster
17-Oct-2006, 01:29
If you were ATI and wanted to blindside an earlier launching G80 by leaking a low number that would have Nvidia thinking they'd got you cold, only to come back with R300 style insanity....?
I don't think NV is the least concerned with R600 being "the new R300".
They know G80 is a very promising design for the future, the DX10 software spec is well know for a long time now, and they built G8x with scalability in mind, like it's SM 3.0 predecessors. And there's always that well oiled marketing machine, of which ATI has consistently fallen victim in the last 2 years, much to market competition's disgust (and consumer anger, due to persistent high prices -longer cicles-).
Let us not forget why was the NV3x 10 months late and underperforming.
If it had come out in early 2002, and with the NV35 refresh in the Fall, history could have been different, no ?
You can't blame it all on NV's decision to throw everything and the kitchen sink into it (both design, DDR2 memories, and manufacturing process), compromising performance in DX9 alone...
Microsoft was pretty mad with NV because of the costly Xbox contract re-negotiations, and decided to punish them by keeping the DX9 API from them until it was too late to change the GPU design ;)
And the fact that ATI was signing contracts at the time for the X360 only made it even more suspitious (the whole "FP24 vs FP16/FP32" DX9 thingy).
Microsoft was pretty mad with NV because of the costly Xbox contract re-negotiations, and decided to punish them by keeping the DX9 API from them until it was too late to change the GPU design ;)
And the fact that ATI was signing contracts at the time for the X360 only made it even more suspitious (the whole "FP24 vs FP16/FP32" DX9 thingy).
Well we don't know if that was a power move by nV or Microsoft :wink: , it could have been from nV or from both sides.
Well we don't know if that was a power move by nV or Microsoft :wink: , it could have been from nV or from both sides.
IIRC Nvidia tried to push their own architecture for DX9 at the design process, with the xbox deal they thought Microsoft would support them but instead it was voted down by everybody. Nvidia ended up leaving, later citing reasons of not wanting to expose some code or IP. They never had the final spec when developing the NV30. Wasn't the partial precision flag added later in the DX9a spec for them?
Sorry no sources, this was a long time ago... Someone correct me if I am wrong.
I really can't understand why 64 pipes has any credibility.
If Orton winked when he said 96 pipes recently, why are people still beating round the bush?
Jawed
Take 96 pipes and use dynamic logic for the ALUs and you have a 600MHz chip with 2-3GHz ALUs. You end up with a smaller die and more performance than G80.
Let us not forget why was the NV3x 10 months late and underperforming.
If it had come out in early 2002, and with the NV35 refresh in the Fall, history could have been different, no ?
You can't blame it all on NV's decision to throw everything and the kitchen sink into it (both design, DDR2 memories, and manufacturing process), compromising performance in DX9 alone...
Microsoft was pretty mad with NV because of the costly Xbox contract re-negotiations, and decided to punish them by keeping the DX9 API from them until it was too late to change the GPU design ;)
And the fact that ATI was signing contracts at the time for the X360 only made it even more suspitious (the whole "FP24 vs FP16/FP32" DX9 thingy).
Perhaps Nvidia just screwed up. Just like ATI dropped the ball with R520. Sometimes things happen without conspiracy.
I believe that Microsoft went with the unified architecture because it just made sense to them.
:lol: Anyone who lived thru "extreme pipes" is not going to forget that experience. :cool:
And just when I was starting to be able to sleep again, you go and bring something like that up.
Xtreme, no?
Take 96 pipes and use dynamic logic for the ALUs and you have a 600MHz chip with 2-3GHz ALUs. You end up with a smaller die and more performance than G80.
I think you'd also burn up a whole lot more power that way...
And the fact that ATI was signing contracts at the time for the X360 only made it even more suspitious (the whole "FP24 vs FP16/FP32" DX9 thingy).
"the whole "FP24 vs FP16/FP32" DX9 thingy" was decided long before Microsoft even started taking bids on Xbox 2.
INKster
17-Oct-2006, 05:10
"the whole "FP24 vs FP16/FP32" DX9 thingy" was decided long before Microsoft even started taking bids on Xbox 2.
Then why didn't they inform Nvidia of that fact in due time ?
Would Nvidia really have made that huge mistake of having FP16 and FP32, but not FP24, if they knew this last mode was to become not only part of the spec, but the de facto standard used by most developers ?
And how come didn't R300 support FP32 also, since it was a part of the spec ?
The only reasonable explanation is that MS gave the upper hand to ATI early, just like it had given it to Nvidia on DX8.1, for instance.
And this is unfair competition interference by MS according to most standards, i think.
That is the reason why i think R600 will not become another R300, just like G80 won't be another NV30.
Microsoft needs compeling reasons for the masses to adopt DX10, since it's part of the "exclusivity" given by Windows Vista. That can't be accomplished when half of the market can't compete with the other, leading to high prices of the hardware, leading to slower DX10 adoption rate ("if DX9 is good enough"...).
DirectX 9.L will be a DirectX 10 for Windows XP DirectX 9.0 L is simply a renamed and refurbished DirectX 10 for Windows XP. It will make DirectX 10 games to work on Windows XP.
And games such as the upcoming Crysis won't work on the existing DirectX 9.0 c. they need a DirectX 9.0 L http://www.theinquirer.net/default.aspx?article=35110
What's up with DX9L for WinXP and DX10 for Vista; what a confusion it is going to be.
ATI and Nvidia needs [totaly/complete] new driver for XP DX9L :( and DX10 for vista.
Why just don't make same:)
What is also amazing when technology becomes advanced/better, so does the price go up.
R300 was origin $399.99
R420 was origin $449.99
R520/R580 was origin $499.99/$549.99
R600 release could be $599.99 or extreme version could be $649.99
Getting very close to intel/AMD for extreme version CPU's $1000 mark...
But ATI/Nvidia probably will...
Chalnoth
17-Oct-2006, 06:08
The whole FP16/FP32 vs. FP24 issue had basically nothing whatsoever to do with the poor performance of the NV3x. The NV30 had fully half of its ALU power in integer-precision, which was entirely useless in DX9. And all the NV3x architectures suffered from an inability for nVidia to make an effective shader compiler. At the release of the NV40, nVidia admitted (I think it was David Kirk, but I'm not certain) that one major problem with the NV3x is that the very last thing that they did was develop the shader compiler. With the NV4x, they developed the shader compiler in conjunction with the core.
The whole FP16/FP32 vs. FP24 issue had basically nothing whatsoever to do with the poor performance of the NV3x. The NV30 had fully half of its ALU power in integer-precision, which was entirely useless in DX9. And all the NV3x architectures suffered from an inability for nVidia to make an effective shader compiler. At the release of the NV40, nVidia admitted (I think it was David Kirk, but I'm not certain) that one major problem with the NV3x is that the very last thing that they did was develop the shader compiler. With the NV4x, they developed the shader compiler in conjunction with the core.
They (nvidia) ended up taking shortcuts on the rendering pipeline with the NV3X, right? I seem to recall them cutting several stages out and implementing some hacky shortcuts to get the job done, albeit at a much slower pace.
:???:
Geeforcer
17-Oct-2006, 07:11
What is also amazing when technology becomes advanced/better, so does the price go up.
R300 was origin $399.99
R420 was origin $449.99
R520/R580 was origin $499.99/$549.99
R600 release could be $599.99 or extreme version could be $649.99
Getting very close to intel/AMD for extreme version CPU's $1000 mark...
But ATI/Nvidia probably will...
If you look at both the performance and R&D cost increases, it doesn't look all that unreasonable.
If you look at both the performance and R&D cost increases, it doesn't look all that unreasonable.
I see were you coming from; you may have a point...
But if you look at the CPU side:
Intel-
Core 2 Extreme (Conroe) 2.93GHz - price $999.99
Core 2 Duo (Conroe) 2.67GHz - price $530.00
Almost twice the money for just about 15% percent increase.
How about between Radeon X1900XTX vs. X1950XTX about $100.00 dollars diffrents for only about 2-7% percent increase.
The_Wolf_Who_Cried_Boy
17-Oct-2006, 10:43
Surround gaming anyone? (http://www.hothardware.com/viewarticle.aspx?page=3&articleid=884&cid=9)
One system that caught our eye was ATI's triple-headed setup running three independent desktops on each with different apps in motion as well.
http://www.hothardware.com/articleimages/item884/big_ati_triplehead.jpg
Well, I assume it's running triple head natively.
Triple head = 1 GPU + an integrated GPU mobo, they did that back with the r420.
Regarding the fp16/24/32 thing, fp32 was never part of dx9.0.
9.0 just had 'minimum fp24' for PS2.0.
What is also amazing when technology becomes advanced/better, so does the price go up.
R300 was origin $399.99
R420 was origin $449.99
R520/R580 was origin $499.99/$549.99
R600 release could be $599.99 or extreme version could be $649.99
Getting very close to intel/AMD for extreme version CPU's $1000 mark...
But ATI/Nvidia probably will...
Since NV40, its Nvidia thats driving the prices higher and higher. Nvidia has already achieved the price point of $649 with the 7800 GTX 512 ..
nutball
17-Oct-2006, 12:30
Since NV40, its Nvidia thats driving the prices higher and higher. Nvidia has already achieved the price point of $649 with the 7800 GTX 512 ..
Since NV40 its NVIDIA who have been launching first. Are these related?
Ailuros
17-Oct-2006, 12:33
Since NV40, its Nvidia thats driving the prices higher and higher. Nvidia has already achieved the price point of $649 with the 7800 GTX 512 ..
Chip complexity, memory prices and demand are the factors that would hypothetically raise a GPU's price and not because any IHV suddenly wakes up in the morning and decides to sell one at X price. You folks are presenting the whole thing as if you've never seen price levels like that. If the Voodoo5 6000 would had launched it would had been priced at what exactly?
As for the example above, the answer is exactly the same as to why a X1950XTX costs roughly a 100 bucks more than a 7900GTX; rare and expensive memory.
There are high end GPUs available at launch of each refresh/generation line that rate in the 300-400$ level and frankly they deliver enough performance to the average high end gamer.
Competition is a fine thing, but it takes at least two to tango. It'd take me a page full to start listing the Ultras, XT, XT-PEs of each side (with or w/o cheese).
R600= 2x Xenos Mother Die+GDDR4 Memory Controller+Avivo 2.0?
trinibwoy
17-Oct-2006, 12:44
R600= 2x Xenos Mother Die+GDDR4 Memory Controller+Avivo 2.0?
Z O M G
Chip complexity, memory prices and demand are the factors that would hypothetically raise a GPU's price and not because any IHV suddenly wakes up in the morning and decides to sell one at X price.You folks are presenting the whole thing as if you've never seen price levels like that. If the Voodoo5 6000 would had launched it would had been priced at what exactly?
7800GTX 512 :!:
As for the example above, the answer is exactly the same as to why a X1950XTX costs roughly a 100 bucks more than a 7900GTX; rare and expensive memory.
X1950XTX and 7900GTX have the same price..
Anyway I was just pointing out the fact that Nvidia has been the first out of the gate since NV40 and they have been setting the prices pretty much. ATI has done everything from offering products at same price points and cheaper even though their GPUs are significantly costlier to produce. I see the same repeating over with G80/R600.
Ailuros
17-Oct-2006, 13:30
7800GTX 512 :!:
Which part of rare and high speced ram is it you don't understand?
X1950XTX and 7900GTX have the same price..
http://www.pricewatch.com/video_cards/992966-1.htm
http://www.pricewatch.com/video_cards/991909-1.htm
Anyway I was just pointing out the fact that Nvidia has been the first out of the gate since NV40 and they have been setting the prices pretty much. ATI has done everything from offering products at same price points and cheaper even though their GPUs are significantly costlier to produce. I see the same repeating over with G80/R600.
Do they have any other choice? The fact that their GPUs cost more to produce (which by the way is valid only for the R5x0 generation) means merely that they make less profit in the end and I'm not in the least interested in the financial figures of either/or company.
It's not only a vast misconception that NVIDIA is driving high end GPU prices higher and ATI supposedly trying to lower them it's also a blantant lie. If there wouldn't be any buyers that are willing to spend 650 or even more bucks neither/nor would release any of those GPUs either.
Which part of rare and high speced ram is it you don't understand?
1.1ns ram was used on the X1900 series as well, it wasnt $100 higher than the competition. You seem to be giving Nvidia a free pass on the GTX512 fiasco .. :runaway:
http://www.pricewatch.com/video_cards/992966-1.htm
http://www.pricewatch.com/video_cards/991909-1.htm
*sigh* (http://anandtech.com/video/showdoc.aspx?i=2858&p=2)
Do they have any other choice? The fact that their GPUs cost more to produce (which by the way is valid only for the R5x0 generation) means merely that they make less profit in the end and I'm not in the least interested in the financial figures of either/or company.
It's not only a vast misconception that NVIDIA is driving high end GPU prices higher and ATI supposedly trying to lower them it's also a blantant lie. If there wouldn't be any buyers that are willing to spend 650 or even more bucks neither/nor would release any of those GPUs either.
There is a gross assumption you are making there, I never meant that ATI was trying to lower the prices. They lowered their prices because they had to not because they wanted to.
You are ignoring my main point; Nvidia has been launching first and essentially setting the price.
Anyways this is all too much offtopic, can we get back on topic?:devious:
http://www.pricewatch.com/video_cards/992966-1.htm
http://www.pricewatch.com/video_cards/991909-1.htmI don't really see how you can say the X1950 is $100 more than the 7900GTX based on those two pages. In fact, I would say they are about equal, all things considered.
John Reynolds
17-Oct-2006, 16:14
And there's no doubt that 7800GTX pricing was higher due to ATI's delays with R520. And I write that with no intended negative connotations attached, since any company would've done the same in the face of missing competition.
Exactly. And it appears that they (ATI that is) are about to hand that same favor to NV again. I get odd looks when I say that hardcore greenies ought to be the most annoyed about R600 coming months later than G80, but it's true. . . they're the ones it's likely to hit in the wallet.
Geeforcer
17-Oct-2006, 17:18
Since NV40, its Nvidia thats driving the prices higher and higher. Nvidia has already achieved the price point of $649 with the 7800 GTX 512 ..
It's supply, demand and all the good stuff that are driving the price up, coupled with execution. If you get your product out on time and the other guy doesn't, you are free to set the price at whatever will maximize your profit, since there is no competetive pressure to check you.
Geeforcer
17-Oct-2006, 17:24
I see were you coming from; you may have a point...
But if you look at the CPU side:
Intel-
Core 2 Extreme (Conroe) 2.93GHz - price $999.99
Core 2 Duo (Conroe) 2.67GHz - price $530.00
Almost twice the money for just about 15% percent increase.
How about between Radeon X1900XTX vs. X1950XTX about $100.00 dollars diffrents for only about 2-7% percent increase.
To me, that's more indicative of how overpriced top-tier CPUs are. Super high-end GPUs have a much better price/performance ratios compared to the products in the lower price ranges than CPUs.
Too many garbage posts. Lets stick to topic.
I think you'd also burn up a whole lot more power that way...
One of the benefits of dynamic logic was lower power consumption. That is why they were able to do 2GHz-2.5GHz at 130nm.
Fast14 technology delivers:
dramatic speed improvements (3-6X) over conventional static design via selective utilization of dynamic logic
a toolset that provides correct-by-construction circuits, early timing convergence, and visibility into physical attributes
timing convergence loop elimination through NDL gate structure and multiphase overlapping clocks
decode function elimination with NDL (1-of-N Dynamic Logic)
designer productivity enhancements via adoption of a semicustom methodology
power and area optimization by redundancy elimination
advanced power reduction techniques (clock gating, null propagation)
noise immunity with Expert Routing Technology
One of the benefits of dynamic logic was lower power consumption. That is why they were able to do 2GHz-2.5GHz at 130nm.Fast14 is a way to get most of the benefits of custom logic at a fraction of the R&D cost. NVIDIA seems to have fully custom designed their ALUs this generation. So I'm not sure what you're trying to get at?
Uttar
trinibwoy
17-Oct-2006, 19:50
Exactly. And it appears that they (ATI that is) are about to hand that same favor to NV again. I get odd looks when I say that hardcore greenies ought to be the most annoyed about R600 coming months later than G80, but it's true. . . they're the ones it's likely to hit in the wallet.
I would think hardcore greenies relish the opportunity to wave their shiny new Geforces in the faces of hardcore reds. Worth a couple bucks I'm sure :grin: The ones most bothered would be the color-blind folks who want the best bang for the buck.
Ailuros
17-Oct-2006, 21:06
1.1ns ram was used on the X1900 series as well, it wasnt $100 higher than the competition. You seem to be giving Nvidia a free pass on the GTX512 fiasco .. :runaway:
Not in the very least. Anyone with a tad of a brain would had stayed away from a 512 back then and would had picked a vendor overclocked G70 instead at way lower prices.
What I'm tired is reading all those years over and over again constant bickering about how expensive GPUs have gotten. Pardon me again but you can get high end GPUs from both IHVs at roughly 200 bucks these days.
There is a gross assumption you are making there, I never meant that ATI was trying to lower the prices. They lowered their prices because they had to not because they wanted to.
Which is for the benefit of the final user yes or no?
You are ignoring my main point; Nvidia has been launching first and essentially setting the price.
And ATI would set the upper price thresholds lower if things would had been different? That's not even the point either; ultra high end GPUs are damn expensive to manufacture at launch, even more so today where performance and chip complexity has increased quite a tad more than in the past. Nothing comes for free and no I don't see how any coming D3D10 GPU with at least half a million transistor and =/>768MB ram can be sold at less than 600 bucks at first if companies really don't want to sell at a loss.
I really can't understand why 64 pipes has any credibility.
If Orton winked when he said 96 pipes recently, why are people still beating round the bush?Whoa, whoa, I asked if anyone who attended the presentation saw him wink, partly in jest, but mostly in desperation for tangible R600 clues. I didn't actually say he did, nor do I know anyone who did. It's possible Damage and the other attendees were too busy scribbling/typing notes to see if he even did wink. :)
That said, 64 shaders sounds weak compared to both Xenos and R580. But that said, if ATI's still carrying a bigger transistor/die area burden b/c of more advanced branching (and possibly other, who knows) features, they could be compensating with super-high clocks, wagering that power draw isn't that big of a deal.
But I believe that one Orton clue may be more precise than any of the nth-hand G80 leaks we've been regurgitating and ruminating on. So, I'm with you, Jawed, if not as forcefully.
As for trumphsiao mentioning in the old thread both that R600 and G80 perform evenly and that R600 is rumored to pull 50-100W more than G80, it seems virtually inconceivable that ATI is so inept as to give up that much power for no performance advantage. We have to assume some performance advantage(s)--and possibly large ones--if R600 pulls that much more power than G80. But that's getting way off-track, especially since we seem to be debating with fewer clues than have been leaked about G80.
I tell you one thing, though. 250 watts'd set that bush on fire, I don't care what double-height cooler they manage to bolt on. :lol:
Not in the very least. Anyone with a tad of a brain would had stayed away from a 512 back then and would had picked a vendor overclocked G70 instead at way lower prices.
What I'm tired is reading all those years over and over again constant bickering about how expensive GPUs have gotten. Pardon me again but you can get high end GPUs from both IHVs at roughly 200 bucks these days.
I'm on the same page as you on being tired of people bickering about prices of high end GPUs.
Which is for the benefit of the final user yes or no?
No. Its for the benefit of themselves; more sales.
And ATI would set the upper price thresholds lower if things would had been different? That's not even the point either; ultra high end GPUs are damn expensive to manufacture at launch, even more so today where performance and chip complexity has increased quite a tad more than in the past. Nothing comes for free and no I don't see how any coming D3D10 GPU with at least half a million transistor and =/>768MB ram can be sold at less than 600 bucks at first if companies really don't want to sell at a loss.
Did I say so? You have gone from doing gross assumptions to putting words in my mouth now. :roll: If R600 would have been launching first, it would be setting the price and G80 following the trend or if Nvidia was strategically postioned well, they could price it even lower. I'm just pointing out the obvious here, ATI was never in a position to set the upper price point in the last two years or so.
What's up with DX9L for WinXP and DX10 for Vista; what a confusion it is going to be.
ATI and Nvidia needs [totaly/complete] new driver for XP DX9L :( and DX10 for vista.
Why just don't make same:)
This doesn't really belong to this thread, but Inquirer is just nothing but plain wrong there.
"DX9.L" aka Direct3D9Ex is a bit enhanced Direct3D9, it has some additions like cross-processing shared surfaces and all the memory resources are managed etc, it was made for Aero GUI's needs, and it's exclusive to Vista, just like Direct3D 10 is.
Bouncing Zabaglione Bros.
17-Oct-2006, 23:14
This doesn't really belong to this thread, but Inquirer is just nothing but plain wrong there.
"DX9.L" aka Direct3D9Ex is a bit enhanced Direct3D9, it has some additions like cross-processing shared surfaces and all the memory resources are managed etc, it was made for Aero GUI's needs, and it's exclusive to Vista, just like Direct3D 10 is.
The Inq have already recanted (http://www.theinquirer.net/default.aspx?article=35140) on that original report of DX9.0L being a way of getting DX10 for XP. Just the usual Faud "report-everything-and-hope-something-is-right" shotgun nonsense approach.
It's supply, demand and all the good stuff that are driving the price up, coupled with execution. If you get your product out on time and the other guy doesn't, you are free to set the price at whatever will maximize your profit, since there is no competetive pressure to check you.
You are correct!
But sometimes if you have no choice (example) ATI R300 Radeon 9500Pro 8 pipeline 128bit memory.
They had to compete against Nvidia, so ATI was not making "good profit" on 9500Pro, they replaced with 9600Pro which only had 4 pipelines.
You can sell less; but it goes against their interest: More profit the better it is...
Then why didn't they inform Nvidia of that fact in due time ?
Would Nvidia really have made that huge mistake of having FP16 and FP32, but not FP24, if they knew this last mode was to become not only part of the spec, but the de facto standard used by most developers ?
And how come didn't R300 support FP32 also, since it was a part of the spec ?
The only reasonable explanation is that MS gave the upper hand to ATI early, just like it had given it to Nvidia on DX8.1, for instance.
And this is unfair competition interference by MS according to most standards, i think.
There are logical explanations for the situation. Nvidia started designing their architecture to support dual precision and ATI wanted a single precision architecture. By the time Microsoft made a decision on the minimum spec it didn't make sense for any significant changes to be made. Plus, as Chalnoth said you can't blame NV30s poor performance solely (or even mostly) on FP24.
Fast14 is a way to get most of the benefits of custom logic at a fraction of the R&D cost. NVIDIA seems to have fully custom designed their ALUs this generation. So I'm not sure what you're trying to get at?
Uttar
Except that fast-14 is custom logic for building dynamic logic circuits and not custom logic like an ALU. Just like both Nvidia and ATI use logic libraries for build CMOS designs.
Fast14 Technology is a unique design technology based on a new logic family (NDL®) that delivers substantial performance, area, and power advantages over conventional CMOS design approaches. Deriving its name from the atomic number of silicon, Fast14 Technology literally translates to mean Fast Silicon Technology. Fast14 Technology does not require exotic manufacturing processes, and the benefits become more pronounced as geometries migrate below 150nm.
See Nvidia is doing nothing new and will not gain anything we haven't seen before. Obviously they have created a staged pipeline architecture like netburst where simple stuff executes faster. If dynamic logic makes a showing in R600 not only will we see logic running at multigigahertz speeds, but we'll see a cooler less power-hungry chip. I find it quite exciting because if they can do 2-2.5GHz at 130nm using a standard TSMC process just think what they will do at 80nm or 65nm.
http://english.www.gov.tw/TaiwanHeadlines/index.jsp?categid=9&recordid=95789
ATI also planned to introduce its next-generation chip R600 late next quarter in order to compete with Nvida's G80 chip. Both chips are also designed according to the 80-nm process, but will likely shrink to be produced under a 65-nm process by yearend.
http://www.cooltechzone.com/Special_Reports/Insider_Series/ATI_R600_Details_Revealed_200604102280/
3. 80/65nm fabrication process, though ATI wouldn’t elaborate on the exact split as well as why they weren’t sticking to just a single fabrication process.
I assume the split will be high end / low end like in the past.
LeStoffer
18-Oct-2006, 11:59
If dynamic logic makes a showing in R600 not only will we see logic running at multigigahertz speeds, but we'll see a cooler less power-hungry chip. I find it quite exciting because if they can do 2-2.5GHz at 130nm using a standard TSMC process just think what they will do at 80nm or 65nm.
Well, you would have to make sure that you can really master that kind of design before putting it to use on such a massive project as the R600 are. Even though ATI might have made some working prototype with the Fast14, I think that they would put it to use in a less complex mass production project first. And you have to ask yourself: if the R600 really uses Fast14, why would they move to the 80nm process at the same time? To much of a risc, methinks...
Well, you would have to make sure that you can really master that kind of design before putting it to use on such a massive project as the R600 are. Even though ATI might have made some working prototype with the Fast14, I think that they would put it to use in a less complex mass production project first. And you have to ask yourself: if the R600 really uses Fast14, why would they move to the 80nm process at the same time? To much of a risc, methinks...
That is a really good point and I agree with what your saying. Although on the other hand why would Nvidia risk increasing their shader clockes to 1.35 GHz?
R6xx to Utilise 80nm and 65nm Processes (http://www.beyond3d.com/forum/showthread.php?t=29612)
Jawed
http://eetimes.com/news/latest/showArticle.jhtml;jsessionid=1CKEAPUHBLTIMQSNDLSCK HA?articleID=193101194
Intrinsity completed a design services deal with graphics processor company ATI (recently acquired by Advanced Micro Devices) which used the Fast14 technology.
Tom's:
We have heard about large cards - like ones 12" long that will require new system chassis designs to hold them - and massive power requirements to make them run.
Why shouldn't the new cards with ATI's R600 require 200-250 watts a piece and the equivalent of a 500 W power supply to run in CrossFire or G80 in SLI?
http://tomshardware.co.uk/2006/10/18/the_new_graphics_uk/page2.html
12"? Yeah, that'd take a new chassis for me, and I have a nice big full tower of recent vintage. I'm remembering when everyone gasped over the 10" X1800XL AIW.
I hope that's from the fever swamps, or at least the non-optimized engineering sample.
INKster
18-Oct-2006, 22:29
Pretty soon we'll need a whole motherboard just for the graphics card(s).
Let's call it "sisterboard" :D
The launch of the R600 is expected to be somewhere around Christmas this year. http://www.cooltechzone.com/Special_Reports/Insider_Series/ATI_R600_Details_Revealed_200604102280/
What really caught my attention is R600 will be sometime this year 2006.
So after all it will not be that far away release time after G80.
So far the possibility could be like this:
Model ATI Radeon 6
Core Code R600
Transistor Count Over 500 million
Manufacturing Process (microns) 0.08
Core Clock 600-650MHz
Vertex Shaders unify
Rendering (Pixel) Pipelines 64
Pixel Shader Processors unify
Texture Mapping Units (TMU) 32
Raster Operator units (ROP) 32
Memory Clock 2.0GHz+ DDR4
DDR Memory Bus 256 or 512-bit
Memory Bandwidth depends GB/s
Ring Bus Memory Controller 512-bit
PCI Express Interface x16
Molex Power Connectors Yes
Max Threads 1024
API Compliancy DX10 Shader Model 4.0
Memory Capacity 1024MB
DemoCoder
19-Oct-2006, 06:26
That site seems to have even less credibility than Inq/Fuad. Just a few stories ago, they were claiming G80: 1) is delayed won't make deadline. 2) is just a G70 with DX10 bolted on. 3) NVidia is "redoing" the chip design and because of this, ATI will beat them to market. Uh huh.
Twice the site has used the phrase "designed from the ground up for DX10" which seems to be a PR talking point, I'm gathering, a strategy upcoming to be used against G80. G80 will be "not truly designed for DX10" whereas R600 will be "built from the ground up" for DX10.
INKster
19-Oct-2006, 06:27
What really caught my attention is R600 will be sometime this year 2006.
So after all it will not be that far away release time after G80.
Did you see the date on that article ? :roll:
Both G80 and R600 have been delayed a bit because of Vista's own delays and other issues, but nothing serious.
Did you see the date on that article ? :roll
What I read was something to think about; I never found a website that would be able to tell exact day+time when a company will release their new technology unless they (ATI) reveal themselves about it...
People are people if they made mistake; we all do, I’ am not saying they are 100% correct but you never know “It could be”
Galduta
19-Oct-2006, 11:29
http://www.digitimes.com/NewsShow/20061019PB200_files/1_r.jpg
The r600 cooler ?:roll: :roll: Is the GeCube 1950 pro (http://www.cdrinfo.com/Sections/News/Details.aspx?NewsId=18596) -
3 PCIe slots, they remember? ;) (http://www.ocworkbench.com/ocwbcgi/newspro/viewnews.cgi?newsid1158301575,93080,) -
http://www.digitimes.com/news/a20061019PB200.html
http://www.digitimes.com/NewsShow/20061019PB200_files/1_r.jpg
The r600 cooler ?:roll: :roll: Is the GeCube 1950 pro (http://www.cdrinfo.com/Sections/News/Details.aspx?NewsId=18596) - 3 PCIe slots ;) (http://www.ocworkbench.com/ocwbcgi/newspro/viewnews.cgi?newsid1158301575,93080,) -
http://www.digitimes.com/news/a20061019PB200.html
Is that a new chopper design for the navy? :shock:
WTF a 3 slot 1950? Why? Did Gecube go off the deep end? :grin:
WTF a 3 slot 1950? Why? Did Gecube go off the deep end? :grin:
it's a overclocked version
it's a overclocked version
Still 3 slot cooler for an original 1 slot design :wink: .
Still 3 slot cooler for an original 1 slot design :wink: .
it's very silent too
The Inq had another blurp about R600. GDDR3 parts codenamed UFO thanx to Samsung screwing up predictions about GDDR4 availability: http://uk.theinquirer.net/?article=35198
http://img222.imageshack.us/img222/117/1wf0.jpg
there you have it :roll: :lol:
http://img222.imageshack.us/img222/117/1wf0.jpg
there you have it :roll: :lol:
Is photoshop a friend of yours?
Is photoshop a friend of yours?
not mine , i'm just the messenger of fun
not mine , i'm just the messenger of fun
shh..don't say it too loudly, Fudo might write a story about it.. :)
Ailuros
19-Oct-2006, 13:05
I want dat separate 20MB memory controller chip thingy now :P
That one musta come from the Rydermark presentation a couple of weeks back :?:
Galduta
19-Oct-2006, 13:32
http://img222.imageshack.us/img222/117/1wf0.jpg
there you have it
http://www.ixbt.com/video2/images/r580/x1900xtx-chip.jpg
:lol: :lol: :lol: :lol: You do not have compassion of us ted.
and sorry for my bad english :sad:
If it was posted before, I apologise for that. It's a news from Digitimes...
ATI schedules 45nm production at TSMC for 2008 (http://digitimes.com/mobos/a20061019PB200.html)
It's interesting to note on the news is here...
Edward Chou, marketing director of ATI Technologies Asia-Pacific division, stated that yield rates on 80nm technology production at Taiwan Semiconductor Manufacturing Company (TSMC) remain stable. ATI has been working closely with the foundry, according to Chou, adding that the company is looking to adopt the 45nm node by 2008. ATI began 80nm production at TSMC at the end of 2005 and started production of its 65nm-made chips from the middle of this year.
Do anyone know what kind of ATi chip that would adopt this 65nm node of the TSMC by the middle of this year?
SugarCoat
19-Oct-2006, 15:27
The Inq had another blurp about R600. GDDR3 parts codenamed UFO thanx to Samsung screwing up predictions about GDDR4 availability: http://uk.theinquirer.net/?article=35198
I wouldnt of expected mainstream parts to have GDDR4 anyway to keep the premium and costs to manufacture low. Those parts dont tend to be terribly bandwidth limited regardless.
Sunrise
19-Oct-2006, 18:21
Do anyone know what kind of ATi chip that would adopt this 65nm node of the TSMC by the middle of this year?
Those should be the test-wafer runs i heard about in April, to qualify their next process nodes for each IHVs needs. Since ATi themselves stated that their next line-up will be a split between 80nm and 65nm production, i think you can guess which ASICs they are referring to.
We got to know that R600 is a 300W beast built on 80nm process technology. To cool off that beast, you need a 4 heatpipes cooling solution. The R600 card is long from what we have seen, around G80GTX length but the cooler is longer than the PCB. It has two power connectors like the NV flagship card and it is dual slot of course. Also we knew that the last spin at TSMC is less than ideal so we can conclude it is suffering from poor yield at this point of time. It is slated to launch in Q1 2007 according to ATi's roadmap but it might delay further.
From VR-Zone (http://www.vr-zone.com/?i=4181)
Kanyamagufa
19-Oct-2006, 19:06
This thing better be the 'Black Pearl' of GPU's...
zsouthboy
19-Oct-2006, 19:09
Question: is the R600 the (cancelled) R400? I remember reading something to that effect back when R420 came out....
From VR-Zone (http://www.vr-zone.com/?i=4181)
Let us recall that the last time VR-Zone opined wisely on ATI yields, it was that the 32-pipe R520 was so darn bad yield-wise that they were only getting 2-3% yields, and weren't even getting enough of the 24-pipe variant to launch with. Of course, the yields of 32/24 pipe R520 were even worse than VR-Zone reported, since there wasn't any. ;)
Just for some historical perspective on VR-Zone and ATI yield information. . . .
. . . they may do reasonably good on NV information from time to time, but they have no street cred with me at all on ATI information.
trinibwoy
19-Oct-2006, 19:29
Does anybody have street cred when it comes to ATI info? As far as I can recall any reliable pre-launch ATI info in the recent past has come out of this forum.
PeterAce
19-Oct-2006, 19:30
Question: is the R600 the (cancelled) R400? I remember reading something to that effect back when R420 came out....
Why not have the answer from the past.... (the 'frying pan wielding' wavey) himself :
http://www.beyond3d.com/forum/showpost.php?p=358401&postcount=34
Let us recall that the last time VR-Zone opined wisely on ATI yields, it was that the 32-pipe R520 was so darn bad yield-wise that they were only getting 2-3% yields, and weren't even getting enough of the 24-pipe variant to launch with. Of course, the yields of 32/24 pipe R520 were even worse than VR-Zone reported, since there wasn't any. ;)
Just for some historical perspective on VR-Zone and ATI yield information. . . .
. . . they may do reasonably good on NV information from time to time, but they have no street cred with me at all on ATI information.
The rumors were right that the R520 was suffering from problems however. Even if the exact details were wrong.
Does anybody have street cred when it comes to ATI info? As far as I can recall any reliable pre-launch ATI info in the recent past has come out of this forum.
Hexus, DailyTech, and AnandTech. TechPowerUP used to score some interesting ppt decks, but they've not nearly as much of late. xbit to some degree too, tho I think they do a bit of guessing too.
Other than beyond3d forum, as noted.
The rumors were right that the R520 was suffering from problems however. Even if the exact details were wrong.
No, the exact details weren't "wrong". The exact details were made up out of whole cloth and had no basis in anything even tangentially based in reality. And yeah, that makes a difference.
Sunrise
19-Oct-2006, 20:29
. . . they may do reasonably good on NV information from time to time, but they have no street cred with me at all on ATI information.
Even "reasonably good" is giving them a little too much credit, as VR-Zone never acts as the messenger itself nor the origin of the information they gather, so they aren´t anymore credible on NV than they are on ATi. It´s the real sources that they depend on that make them look like they are credible or not.
INQ is doing a lot of the same, but additionally, they even try to imply that they really know what they are talking about, while VR-Zone at least keeps away from doing that too often and that´s also the reason why they are useful to us, because their staff is at least able to read those damn chinese/asian language slides, which is also why they are quoted constantly, because they write it down in English.
E.x. those 24/32pipe R520 rumours were a copy of HKEPC, their G80 info is actually coming from the posted chinese slides (they scored on that one)...the list just goes on and on.
Site like, e.x. Dailytech, do a much better job recently and they also have done that in the past.
Could this be the test probe for cooling solution that R600 will use?
http://www.gecube.com/cms_file.php?id=2054
It seems a little strange that ATI would dump the byproduct of 200-300W in the case rather than try to exhaust some of it, considering they do so now with considerably lower power draw.
INKster
19-Oct-2006, 22:51
The vr-zone newsbit indeed says "4 heatpipes", but they also say that it is dual-slot and longer than the PCB, so this can't be it.
Besides, like Pete said above, dumping heat inside the case is like going back 2/3 years in time, it makes no sense for a high-end card.
Skrying
19-Oct-2006, 22:57
First, if I can get my hands on that GeCube X1950 Pro then it'll be the one I'm picking up!
Besides that. I personally expect the heatsink and fan to be similair to that of the new X1950XTX HSF. But with heat pipes that wrap around to a heatsink on the back of the card, I also expect a slightly larger and redesigned fan to be used.
The upcoming GF8800 line will hit just under 12,000 in 3DMark06. http://www.theinquirer.net/default.aspx?article=35195
Well, I will not say if the Inq are wrong or right.
If two Geforce 7900GTX SLI scores 10k in 3DMark06 - and single 8800GTX will be 11.5k in 3Dmark06 I would say it is prety good. "The only problem is power requirement would be too high"
Also how the Inq be so sure if R600 will be better then G80?
vertex_shader
20-Oct-2006, 10:39
Let us recall that the last time VR-Zone opined wisely on ATI yields, it was that the 32-pipe R520 was so darn bad yield-wise that they were only getting 2-3% yields, and weren't even getting enough of the 24-pipe variant to launch with. Of course, the yields of 32/24 pipe R520 were even worse than VR-Zone reported, since there wasn't any. ;)
Just for some historical perspective on VR-Zone and ATI yield information. . . .
. . . they may do reasonably good on NV information from time to time, but they have no street cred with me at all on ATI information.
Solution is easy, Vr-Zone=Shamino=nv "playing pet", ati have "playing pet's too" macci and sampsa, but i never seen macci or sampsa write anything about g80 yield :wink:
I not say this rumor is not true, but for me vr-zone lost all credit in the r520 saga.
Next time i think we will see this rumor on inqurier :wink:
From VRZone (http://www.vr-zone.com/?i=4188)
http://resources.vr-zone.com//newspics/Oct06/20/RV670.gif
Looks like guesses based on what we know so far. :lol: Although they are naive to think that ATI will go back to a 3 SKU stack.
That RV630 --> RV660 & RV670 is a blinking neon sign "WE'RE COMPLETELY GUESSING BASED ON THIS YEAR. ALL WE DID WAS REPLACE THE '5' WITH A '6'. NO ACTUAL INFORMATION WAS LEAKED WITH THIS POST."
Specifications of the G80 chip are not clear. Some sources indicate that Nvidia’s first DirectX 10 chip will incorporate 48 pixel shader processors and an unknown number of vertex shader/geometry shader processors. Other, however, claims that the G80 has 32 pixel and 16 vertex and geometry shader processors. Yet another source has indicated that the G80 will have unified shader architecture and will consist of 700 million transistors.
http://www.xbitlabs.com/news/video/display/20061023142704.html
Asustek Computer and Nvidia Corp. did not comment on the news-story.
I'm pretty sure ATI is also waiting to see how Nvidia G80 will turn out. So ATI then will have a basic idea what they need to do w/R600 in order to make R600 better/faster.
It is uncertain why Asustek and Nvidia use GDDR3 memory for the new top-of-the-range products, as GDDR3 memory operating at 1.70GHz and above is pretty rare, whereas GDDR4 memory operates at 2.0GHz and beyond easily. "
I was wondering same thing; people who over-clock their video cards may have trouble since GDDR3 has reach its limits.
I'm pretty sure ATI is also waiting to see how Nvidia G80 will turn out. So ATI then will have a basic idea what they need to do w/R600 in order to make R600 better/faster.
I think the only thing they can do with R600 is adjust the clocks at this point. Unless it's a year away :P
Bouncing Zabaglione Bros.
25-Oct-2006, 09:27
I'm pretty sure ATI is also waiting to see how Nvidia G80 will turn out. So ATI then will have a basic idea what they need to do w/R600 in order to make R600 better/faster.
ATI is pretty limited at this point with regards to making changes to combat G80. It's not like they can suddenly throw in another 48 unified shaders or anything more significant in the chip. They might be able to fiddle with the clocks, or throw massive amounts of money away by having only high yielding products in their range (ie having a product range that consists of the higher specced bins), but I doubt that will happen.
Geeforcer
25-Oct-2006, 09:43
ATI is pretty limited at this point with regards to making changes to combat G80. It's not like they can suddenly throw in another 48 unified shaders or anything more significant in the chip. They might be able to fiddle with the clocks, or throw massive amounts of money away by having only high yielding products in their range (ie having a product range that consists of the higher specced bins), but I doubt that will happen.
Another (costly) thing they could do is use faster memory. But yeah, outside of chip and memory clocks, R600 is set.
Another (costly) thing they could do is use faster memory. But yeah, outside of chip and memory clocks, R600 is set.
They could try to completely undercut the G80 on a price perspective.
INKster
26-Oct-2006, 01:51
They could try to completely undercut the G80 on a price perspective.
What's stopping NV from doing the same when the R600 comes out ?
What's stopping NV from doing the same when the R600 comes out ?
Nothing. And I sure do hope the two sides get into a massive pricing war (downward).
ATI is pretty limited at this point with regards to making changes to combat G80. It's not like they can suddenly throw in another 48 unified shaders or anything more significant in the chip. They might be able to fiddle with the clocks, or throw massive amounts of money away by having only high yielding products in their range (ie having a product range that consists of the higher specced bins), but I doubt that will happen.
I remember the days when R420 as oring suppose to be 12 pipeline, but since NV40 was 16 pipeline then ATi squeze extra 4 pipeline to make 16 also.
Well ATI did knock up R420 by bolting two R350s together and adding 3Dc and some twiddly bits :lol:
http://www.beyond3d.com/forum/showpost.php?p=680301&postcount=2
Jawed
I remember the days when R420 as oring suppose to be 12 pipeline, but since NV40 was 16 pipeline then ATi squeze extra 4 pipeline to make 16 also.
This was already planned for, well was it planned for or were they redunent pipes for yeild reason:wink:, and for the refresh they could have gotten the full 16 pipes out with no problem at all.
INKster
26-Oct-2006, 02:17
Well ATI did knock up R420 by bolting two R350s together and adding 3Dc and some twiddly bits :lol:
http://www.beyond3d.com/forum/showpost.php?p=680301&postcount=2
Jawed
One thing is designing 2 x "more of the same".
Another is waiting for the manufacturing process/cost to catch up (not to mention taming power consumption/heat to acceptable levels)...
One thing is designing 2 x "more of the same".
Another is waiting for the manufacturing process/cost to catch up (not to mention taming power consumption/heat to acceptable levels)...
Well it all depends if ATi planned for it in advance they possibly can make something that demolishes the g80, they must have learned something from the past (gf6 with 16 pipes, gf7 with its super drivers and to a lesser extent the 7800 gtx 512, nV seems to always have something ready just in case)
Dave Baumann
26-Oct-2006, 02:25
So, they have a CPU ready? :razz:
So, they have a CPU ready? :razz:
LOL its gotta be a GPU/CPU hybrid nV won't stop in the middle :lol:
INKster
26-Oct-2006, 02:34
So, they have a CPU ready? :razz:
Maybe they can continue to inflate their stock... and then buy AMD in a few years.
It's not like the difference is that big right now. :razz:
Maybe they can continue to inflate their stock... and then buy AMD in a few years.
It's not like the difference is that big right now. :razz:
Dream on! :lol:
INKster
26-Oct-2006, 02:49
Dream on! :lol:
12 months ago AMD was also dreaming with buying ATI, and then look what happened. :D
Jen Hsun is a good friend of Goldman Sachs', maybe they can loan them the money, the same way AMD had half of ATI on a loan by Morgan Stanley Senior Funding.
12 months ago AMD was also dreaming with buying ATI, and then look what happened. :DJen Hsun is a good friend of Goldman Sachs', maybe they can loan them the money, the same way AMD had half of ATI on a loan by Morgan Stanley Senior Funding.
AMD needed ATI, its no compulsion for Nvidia to join the merger or buyout race. They are mighty fine on their own. :cool2:
Back on topic at hand, has anybody approximated the die size of Xenos? Also Dave noted this in his Xenos article:
When we factor in the savings for both power and die size savings we can see that this potentially has some advantages over traditional architectures. In the case of the XBOX 360 not only does this result in a relatively smaller die size for a fairly high performance ratio but also means that the graphics need only be air cooled, without the use of its own additional fan. Beyond the immediate application we can see that unified designs that are bound for the PC could have smaller die sizes for equivalent performances as current discrete solutions or more silicon dedicated to either more ALU's for higher performance, or other transistors dedicated to other functionality.
This was already planned for, well was it planned for or were they redunent pipes for yeild reason:wink:, and for the refresh they could have gotten the full 16 pipes out with no problem at all.
Before we talk; we do not have exact truth about G80 until Nov 8, 2006, when Nvidia will introduce G80. After that only ATI will know if their chip R600; how well its going to compete: But we can only guess.
I moved the majority of the CPU/GPU talk into its own thread. So please continue discussing R600 (and/or R6xx!) here instead, mkay? :)
Uttar
Before we talk; we do not have exact truth about G80 until Nov 8, 2006, when Nvidia will introduce G80. After that only ATI will know if their chip R600; how well its going to compete: But we can only guess.
Sure about the g80? :lol: .
True only ATi knows what the r600 has right now, but if the r600 is performing lower then the g80 as it stands ATi will try to up the clocks and us faster memory to edge out the g80, if the r600 is already faster, well then no changes. Either way it will probably end up close or a slight win for ATi. Well unless the r600 has major issues.
Chalnoth
26-Oct-2006, 16:22
Why should the R600 pull ahead? nVidia might just themselves release a new high-end version soon after if ATI pulls ahead just with a clockspeed bump.
It depends, ATi, if they plan for a speed bump from nV now has a small advantage of coming out after, but then again, we don't know if nV has already made plans for a 80 nm g80. Its all down to business tactics which we don't and won't have a clear understanding and how it will be played out until around the time the r600 is scheduled to be released.
Chalnoth
26-Oct-2006, 16:42
The G80 architecture is sure to last a good two to three years. So they may or may not skip 80nm. But does that really matter? I think we've learned that it seems that today it's overall design that is more important than manufacturing process. Using a more advanced manufacturing process can help, in that it typically will allow for better power consumption, or it can hurt, in that new manufacturing processes typically take some time to get yields to the point where they overtake older manufacturing processes for large dies. At best, using 80nm is going to be a minor influence on the performance of the R600.
vertex_shader
26-Oct-2006, 19:44
Why should the R600 pull ahead? nVidia might just themselves release a new high-end version soon after if ATI pulls ahead just with a clockspeed bump.
Golden classics again....
I see this the topic future:
"r600 coming only in april" "r600 hot and long" "r600 yield 8%" "r600 20% slower than g80" "when r600 coming in january, nv already release a 8800 ultra" "r600 die huge, this is kill ATi margins" "r600 = nv30" "r600 need 600watt psu, r600 cf need 900watt psu" :roll:
Why should the R600 pull ahead? nVidia might just themselves release a new high-end version soon after if ATI pulls ahead just with a clockspeed bump.
I wonder how soon they can be ready with a redesign in order to keep up with R600. It is a good thing they will have a bit of a lead because otherwise how would they compete. Perhaps they have tricks up their sleeve and can put two 8800 GPUs on a card to compete with R600.
DudeMiester
27-Oct-2006, 04:53
Golden classics again....
I see this the topic future:
"r600 coming only in april" "r600 hot and long" "r600 yield 8%" "r600 20% slower than g80" "when r600 coming in january, nv already release a 8800 ultra" "r600 die huge, this is kill ATi margins" "r600 = nv30" "r600 need 600watt psu, r600 cf need 900watt psu" :roll:
Just no. :lol:
Chalnoth
27-Oct-2006, 05:11
I wonder how soon they can be ready with a redesign in order to keep up with R600. It is a good thing they will have a bit of a lead because otherwise how would they compete. Perhaps they have tricks up their sleeve and can put two 8800 GPUs on a card to compete with R600.
Well, I wasn't suggesting that. Just that if the R600 only pulls ahead because ATI sacrifices on yields compared to what they were originally planning to do, releasing a product with higher clockspeeds, then nVidia could do the same right back.
The real differentiator is going to be the particular architectural design choices each company made some time ago. And as far as that is concerned, we're going to have to see which one pulls ahead. If the rumors are correct, it certainly seems like nVidia has done something very radical with the G80 design. What is the R600 going to be like? It will be a very interesting comparison, I think.
http://www.freshpatents.com/Display-screen-subsection-rendering-apparatus-and-method-dt20060316ptan20060055701.php
A method and apparatus for providing rendering of subsections of screen space receives render commands associated with different screen subsections, such as from a command buffer populated by a coprocessor, and determines which screen section is currently being rendered by a rendering engine, or stated another way, which screen section the host processor wishes to have rendered, and evaluates screen subsection data that is associated with a received rendering command. The screen subsection data identifies a screen subsection for which the command refers. The method includes executing the command if it is determined that the command refers to a current screen section being rendered.
Vista requirement / crossfire or something else?
http://www.freshpatents.com/Apparatus-with-redundant-circuitry-and-method-therefor-dt20060309ptan20060053188.php
An apparatus with circuit redundancy includes a set of parallel arithmetic logic units (ALUs), a redundant parallel ALU, input data shifting logic that is coupled to the set of parallel ALUs and that is operatively coupled to the redundant parallel ALU. The input data shifting logic shifts input data for a defective ALU, in a first direction, to a neighboring ALU in the set. When the neighboring ALU is the last or end ALU in the set, the shifting logic continues to shift the input data for the end ALU that is not defective, to the redundant parallel ALU. The redundant parallel ALU then operates for the defective ALU. Output data shifting logic is coupled to an output of the parallel redundant ALU and all other ALU outputs to shift the output data in a second and opposite direction than the input shifting logic, to realign output of data for continued processing, including for storage or for further processing by other circuitry.
Here is how they are handling a bad ALU. You have one spare for a bank of ALUs and if one is bad they all shift data over and utilize the extra ALU.
http://www.freshpatents.com/Graphics-processing-logic-with-variable-arithmetic-logic-unit-control-and-method-therefor-dt20060309ptan20060053189.php
Briefly, graphics data processing logic includes a plurality of parallel arithmetic logic units (ALUs), such as floating point processors or any other suitable logic, that operate as a vector processor on at least one of pixel data and vertex data (or both) and a programmable storage element that contains data representing which of the plurality of arithmetic logic units are not to receive data for processing. The graphics data processing logic also includes parallel ALU data packing logic that is operatively coupled to the plurality of arithmetic logic processing units and to the programmable storage element to pack data only for the plurality of arithmetic logic units identified by the data in the programmable storage element as being enabled.
Power Saver functionality for defective ALU
http://www.freshpatents.com/Method-and-apparatus-for-a-stacked-die-configuration-dt20060309ptan20060051912.php
A stacked die configuration for use in an IC package includes a first IC die mechanically coupled to a substrate material. Mechanically coupled to the first IC die is an interposer having an aperture adapted to receive a second IC die. Mechanically coupled to the first IC die and fitting within the aperture of the interposer is a second IC die. As a result, both the overall height of the IC package and the length of the bond wires connecting each of the members of the stacked die configuration may be reduced.
Anybody know if this sounds like R600 or is this Xbox related? Multiple dies within the same package.
http://www.freshpatents.com/Display-screen-subsection-rendering-apparatus-and-method-dt20060316ptan20060055701.php
Vista requirement / crossfire or something else?
Just seems like a possible minor update to their usual screen tiling approach to rendering.
After success with R300; ATI was working next big step R400 but then later it was cancel with later design of R500 - I believe" which became available to Xbox360. But instead they improved R300 and made R420.
I believe R520/R580 is just simplify apposite world direction of R500 for Xbox360. But R600 instead should be based of R500 for Xbox360 - but improved and design towards different approach in PC based system world.
Second: ATI will not make same mistake twice as example with R520 based 90nm jump. It would be unwise for R600 to follow same error steps and jump directly to 65nm right now (I don't think they could) it make sense to stay with more mature 90nm but slightly move to 80nm which is not that danger.
ATI made some money with Xbox360; so now ATI is free and can spend more time and sources availability to improve R500.
Whatever the reason R600 is delay after G80, it means it is more advance then G80 or ATI has screw-up with R600.
Whatever the reason R600 is delay after G80, it means it is more advance then G80 or ATI has screw-up with R600.
Or their manufacturing cycle is diffent then Nvidia's.
Just seems like a possible minor update to their usual screen tiling approach to rendering.
Perhaps but this feature caught my eye....
...predicated rendering....
So that leads me to believe this is R600 and DX10. Looks more like the ability to render different application windows.
So that leads me to believe this is R600 and DX10. Looks more like the ability to render different application windows.
Agreed. Only the CPU can identify regions of the screen at the time a packet of work is created and a windowing system is the most likely scenario in which screen regions are known in advance.
Jawed
http://www.freshpatents.com/Display-screen-subsection-rendering-apparatus-and-method-dt20060316ptan20060055701.php
Vista requirement / crossfire or something else?
Sounds like Xenos predicated tile rendering for AA to me.
I moved the posts regarding speculations of AMD quitting the high-end market in the 3D Industry forum, same thread that already existed on the subject. Hopefully this thread will remain relatively active even without that, we don't want to be seen as a NVIDIA fan forum, damnit! ;)
Uttar
trinibwoy
28-Oct-2006, 22:54
Well given that it's an R600 thread maybe we could do with some more R600 rumours :)
Chalnoth
28-Oct-2006, 23:08
I moved the posts regarding speculations of AMD quitting the high-end market in the 3D Industry forum, same thread that already existed on the subject. Hopefully this thread will remain relatively active even without that, we don't want to be seen as a NVIDIA fan forum, damnit! ;)
Well, the R600 is still quite a ways off. I'm sure the R600 rumors will pick up when it's closer to release :)
Albuquerque
31-Oct-2006, 21:38
It'll come with a billion transistors operating at 1ghz, a gig of GDDR4 ram at 2ghz, can raster 32 billion pixels per second and can calculate 128 billion 16bit floating point ops per second or 256 billion 16bit integer ops per second.
Requires a one kilowatt powersupply minimum and exhales something around 15,000 BTU's...
:shock: :lol:
I have no clue, I just wanted to nudge the thread a bit...
http://patft1.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PALL&p=1&u=%2Fnetahtml%2FPTO%2Fsrchnum.htm&r=1&f=G&l=50&s1=7,123,266.PN.&OS=PN/7,123,266&RS=PN/7,123,266
Method and apparatus for a video graphics circuit having parallel pixel processing
Generally, the present invention provides a method and apparatus for parallel processing of pixel information within a video graphics circuit. This may be accomplished when the video graphics circuit includes a set-up engine, an edgewalker circuit, a span processing circuit, and a plurality of pixel circuits. In such an embodiment, the set-up engine receives vertex information and produces object-element information therefrom. The object-element information is provided to the edgewalker circuit, which in turn produces span definition information. The span definition information identifies the starting pixel of a span and the starting pixel parameters. The span information is received by the processing circuit and converted into a plurality of pixel processing circuits wherein each of pixel parameters are provided to the plurality of pixel processing circuits wherein each of the plurality of pixel processing circuits processes corresponding pixel parameters to produce pixel information in accordance with the information provided by the processing circuit. With such a method and apparatus, the benefits of parallel processing are achieved in a video graphics circuit without the requirement of doubling, or near doubling, the hardware requirements and without the additional circuitry to track the efficiency and validity of the information being processed.
Looks like the setup engine has been beefed in order to feed the gpu all the vertex information.
hmm interesting find, but this is an old patent :wink:, well it could be still for the r600.
Your right. I was checking the date in the top right corner. Applied for two years before the unified shader patent. I doubt this is R600 related. Weird that it was posted so recently.
well black box patents aren't released to general public until a certain date, maybe ATi felt they had use a black box patent or the attorney that filled it felt it was necessary, they always look for extra money any which way :grin:.
I think they are called black box, our attroney tried to do that for some of our engine code, which we meekly turned down lol, strong believer in can't patent math.
http://www.theinquirer.net/default.aspx?article=35483
Nvidia has a really strong line-up for upcoming Yuletide shopping madness. However, within the ranks of Graphzilla's troopers there is an obvious intent to bury all of the more advanced features that the competition will offer in couple of months' time. 512-bit memory interface, more pixels per clock, second gen RingBus marchitecture... all this is hidden in the dungeons of Markham and DAAMIT's R&D Labs in Santa Clara and Marlboro.
So more then 24 ROPS?
Heh, gotta love the inq, they are so vague they can almost mean anything :lol: , shameless!
INKster
03-Nov-2006, 01:01
Man, i wish this 512bit external bus "thingy" would just go away, it just doesn't make any sense right now.
Out of the blue, ATI would pair an advanced architecture with brute-force methods ? It's not their style. :D
And what about PCB complexity ?
I can't picture a 80nm/65nm GPU surrounded by all sides with memory chips, especially of the top GDDR4 kind.
G80's -90nm, therefore, a large chip- PCB is already insanely complex, and the only side without memory chips (and is "only" GDDR3) is the bottom one, where the PCI-Express lanes lead to the connector.
Is that much memory bandwidth really needed at this point ?
Man, i wish this 512bit external bus "thingy" would just go away, it just doesn't make any sense right now.
Out of the blue, ATI would pair an advanced architecture with brute-force methods ? It's not their style. :D
And what about PCB complexity ?
I can't picture a 80nm/65nm GPU surrounded by all sides with memory chips, especially of the top GDDR4 kind.
G80's -90nm, therefore, a large chip- PCB is already insanely complex, and the only side without memory chips (and is "only" GDDR3) is the bottom one, where the PCI-Express lanes lead to the connector.
Is that much memory bandwidth really needed at this point ?
If ATi is using gddr4 with a 512 bit bus, thats like what 160Gb/sec bandwidth? The r600 better be 3 or more times faster then the r580 to utilize that much bandwidth........
Its just doesn't make any sense to go with a 512 bit bus.
Rangers
03-Nov-2006, 01:41
If ATi is using gddr4 with a 512 bit bus, thats like what 160Gb/sec bandwidth? The r600 better be 3 or more times faster then the r580 to utilize that much bandwidth........
Its just doesn't make any sense to go with a 512 bit bus.
You could probably get 2X performance out of R580 just by un-texture limiting it (it's sitting on fourty eight pixel shader pipes after all).
So yeah, easily possible.
You could probably get 2X performance out of R580 just by un-texture limiting it (it's sitting on fourty eight pixel shader pipes after all).
So yeah, easily possible.
nah maybe 50% more but thats about it.
trinibwoy
03-Nov-2006, 01:45
You could probably get 2X performance out of R580 just by un-texture limiting it (it's sitting on fourty eight pixel shader pipes after all).
So yeah, easily possible.
2X? Uh, no. That would imply that the ALU's are spending half their time spinning their wheels waiting on texture reads.
Rangers
03-Nov-2006, 01:51
2X? Uh, no. That would imply that the ALU's are spending half their time spinning their wheels waiting on texture reads.
Who says they aren't?
Websites have done ATI R520 pipe vs Nvidia G71 pipe clocked the same before (http://www.driverheaven.net/articles/efficiency/), and they come out within 10-20% of each other. You're right, it might not be double, but it might be 1.6-1.8X
trinibwoy
03-Nov-2006, 01:58
Who says they aren't?
Websites have done ATI R520 pipe vs Nvidia G71 pipe clocked the same before (http://www.driverheaven.net/articles/efficiency/), and they come out within 10-20% of each other. You're right, it might not be double, but it might be 1.6-1.8X
G71 pipe = R520 pipe ? Let me put it this way, do you think ATi would come up with a design where most of their functional units are 50% utilized on average?
Rangers
03-Nov-2006, 06:54
G71 pipe = R520 pipe ? Let me put it this way, do you think ATi would come up with a design where most of their functional units are 50% utilized on average?
Well, they did.
Did you read the link? They downclocked a R520, disabled a quad on G71 and compared pipe for pipe clock for clock. ATI lost but not by much. Disregard the opengl game imo. The DX9 and synthetic tests they did well. You could look at a X1800GTO vs 7600GT for more proof.
ATI surely bet on the 3:1 ratio and/or had no choice.
I do agree though, that it's probably more like 60-70% utilized, and they hoped for more shader intensive games to get even more. It was the only way for them to scale easily, apparantly. Sure, tripling the shader units might only gain them 5-10% over just doubling them, but it didn't take a ton of die area so they probably figured they might as well. For some reason apparantly they couldn't scale the texture units. And or apparantly there is something that limits them to either 1:1 or 3:1.
Anyways I guess this is OT. I sure hope they learned a lesson and R600 has about 64 TMU's though, like INQ's hint of "64 true pipes". It seems now that Nvidia has gone to dedicated texture units, they are surely making sure it does not limit them (64 in G80 according to inq?)
NocturnDragon
03-Nov-2006, 11:54
If ATi is using gddr4 with a 512 bit bus, thats like what 160Gb/sec bandwidth? The r600 better be 3 or more times faster then the r580 to utilize that much bandwidth........
It would fit with my signature :lol:
LOL, my thoughts exactly! :lol:
will be fully DirectX 10 compliant, utilising a Unified Shader Model architecture.
R600 will feature 64 Shader pipelines (processing both vertices and pixels) with 32 TMUs and 32 ROPs
R600 is expected to interface to 512MB of 2Ghz+ GDDR4 Memory over a 512-bit interface.
R600 is reported to consume up to 250W (twice that of R580) and may require two PCI Express power connectors.
Boy! this is high expectation's....
G80-384bit memory vs. R600 512bit memory
G80-24 ROP vs. R600 32 ROP
"I think" G80 24 TMUs vs. R600 32 TMU's
G80 32 Shader pipelines vs. R600 64 shaders pipelines
G80 up to 128 shaders enable vs. R600 ????
G80 GDDR3 vs. R600 GDDR4
G80 90nm vs. R600 80nm
G80 partial Unified Shader Architecture vs. R600 Full Unified Shader architecture.
Based on this compare R600 might destroy G80.
Boy! this is high expectation's....
G80-384bit memory vs. R600 512bit memory
G80-24 ROP vs. R600 32 ROP
"I think" G80 24 TMUs vs. R600 32 TMU's
G80 32 Shader pipelines vs. R600 64 shaders pipelines
G80 up to 128 shaders enable vs. R600 ????
G80 GDDR3 vs. R600 GDDR4
G80 90nm vs. R600 80nm
G80 partial Unified Shader Architecture vs. R600 Full Unified Shader architecture.
Based on this comparision R600 might destroy G80.
Both companies have similiar restrictions when it comes to manufacturing the PCB, GPU, and Fab technologies, does it sound reasonable when both companies have similiar reasoruces, skills, and abilities that one would be able to out do the other in reasources, the one thing they don't have any control over?
INKster
04-Nov-2006, 02:44
Something that is "partially unified" isn't unified at all... ;)
It's too early to call R600 a winner. We don't know how it stands against G80 GTX, and changing a bad design is not this trivial task that can be accomplished in a couple of months, between both releases.
Both companies have similiar restrictions when it comes to manufacturing the PCB, GPU, and Fab technologies, does it sound reasonable when both companies have similiar reasoruces, skills, and abilities that one would be able to out do the other in reasources, the one thing they don't have any control over?
Please explain more in detail
INKster
04-Nov-2006, 02:52
Please explain more in detail
He means that both ATI and Nvidia use the same Fab at TSMC, have certain power consumption, thermal and electric crosstalk limits, and both share PCB design and size limitations, memory chip types and speeds, etc.
There is only so much that the design differences can do for a GPU.
He means that both ATI and Nvidia use the same Fab at TSMC, have certain power consumption, thermal and electric crosstalk limits, and both share PCB design and size limitations, memory chip types and speeds, etc.
There is only so much that the design differences can do for a GPU.
Do you think in near future limitation will extended further/better directions since ATI/AMD join and ATI will have access to AMD resources/fab.
Geeforcer
04-Nov-2006, 03:00
Boy! this is high expectation's....
G80-384bit memory vs. R600 512bit memory
G80-24 ROP vs. R600 32 ROP
"I think" G80 24 TMUs vs. R600 32 TMU's
G80 32 Shader pipelines vs. R600 64 shaders pipelines
G80 up to 128 shaders enable vs. R600 ????
G80 GDDR3 vs. R600 GDDR4
G80 90nm vs. R600 80nm
G80 partial Unified Shader Architecture vs. R600 Full Unified Shader architecture.
Based on this compare R600 might destroy G80.
Nevermind the mystical R600 specks, where did you get the bolded parts from?
Nevermind the mystical R600 specks, where did you get the bolded parts from?
Most of it from the Inq.....
G80-384bit memory vs. R600 512bit memory
G80 GDDR3 vs. R600 GDDR4
G80-24 ROP vs. R600 32 ROP
I've reorganized your data a bit, hope you don't mind.
I think 512b is a (possibly) very interesting move on ATI's part. I look forward to seeing what they do with that bandwidth.
"I think" G80 24 TMUs vs. R600 32 TMU's
G80 32 Shader pipelines vs. R600 64 shaders pipelines
G80 up to 128 shaders enable vs. R600 ????
G80 has, apparently, 64 TMUs, with 32 addressing units -- so TMU utilization isn't maximized until you turn the AF up. G80 is running at a rather slow pace, though, compared to where R600 might wind up at....
The G80 doesn't have 32 shader pipelines. You might say it has eight arrays of 16 streaming units each, where a unit has a full MAD and a MUL ALU (SF is in there somewhere, we don't know where yet -- that is, anyone who is likely to comment over the next four days or so doesn't know yet ;) ). Point is, though, it is even less clear what an R600 "shader" looks like. You need to compare not only number, but capability, SIMD-width, speed of calculation, etc. I expect, though, that ATI would win the math contest, given their historical interest in ALU capability.
G80 90nm vs. R600 80nm
I think this gets to the issue of core speed, noise, and cost, but isn't itself otherwise interesting.
G80 partial Unified Shader Architecture vs. R600 Full Unified Shader architecture.
In what way?
Based on this compare R600 might destroy G80.
I would love that to be true! I love leapfrog -- competition is good for technological advancement. I don't expect complete destruction, but, we'll see.
In what way?
I may quoted wrong/not clear.
What I read about G80 - is that actual architecture structure is somewhat different that R500 for Xbox360 desigh how it utilize unifying pixel/vertex/etc routing in certainer scenes in parts of the games.
Skrying
04-Nov-2006, 03:33
Both companies have similiar restrictions when it comes to manufacturing the PCB, GPU, and Fab technologies, does it sound reasonable when both companies have similiar reasoruces, skills, and abilities that one would be able to out do the other in reasources, the one thing they don't have any control over?
True, they have limits, but who in the heck is saying either one is hitting those? Nvidia is certainly not with their current cards while ATi is more to it. Once again Nvidia could be conservative to maximize profit and ATi goes all out.
trinibwoy
04-Nov-2006, 04:37
This is going to be an interesting one. I wonder when ATI found out that G80 is what it is. Nvidia may be bringing a much bigger IQ and efficiency stick to this fight than previously expected. And 681 million transistors on 80/90nm is certainly not comparable to the conservative G71.
True, they have limits, but who in the heck is saying either one is hitting those? Nvidia is certainly not with their current cards while ATi is more to it. Once again Nvidia could be conservative to maximize profit and ATi goes all out.
Neither nV nor ATi have been conservative when it comes ot Fab processes in the past few generations. They know they can't otherwise the other would have an advantage. Look at the clock speed variations of the high end. Both IHV's have been very close, this isn't by accident. Where nV went low was the features, this is something Dx10 doesn't allow anymore, so does ATi has better engineers and knows how to maximize TSMC's Fabrication process?
True, they have limits, but who in the heck is saying either one is hitting those? Nvidia is certainly not with their current cards while ATi is more to it. Once again Nvidia could be conservative to maximize profit and ATi goes all out.
With ATI pursuing a 3:1 ALU vs. texture ratio, and with recent rumors hinting at 32 TMUs, presumably we are expecting 32 TMUs and 96 shader units. Given that R600 is likely 80nm, and with fewer TMUs, ATI has more room to put some weight on those shaders, so I wouldn't be surprised to see a Vec4 + scalar. Call it 9 ops, guess at a 700Mhz clock, and we get 604gflops or so, which would be quite impressive indeed.
Of course, that's all guesswork/fantasy.
Another wild card is whether either one has an fp24/nao32-style efficiency shortcut up their sleeves.
^eMpTy^
04-Nov-2006, 07:18
R600 will have about 40 million more transistors than G80. But if we take into account that secondary I/O chip, the overall count may be higher for G80. Also, if the X1950Pro is any indication, 80nm technology is more about lowering power consumption than it is increasing clock speeds (90nm cards overclock better). So I don't think R600 will gain any ground on that front.
I think this is all going to come down to pipeline layout and how the first dx10 games utilize resources. Given that the GPUs are becoming more general, and dx10 is making specifications more strict, I tend to think that R600 and G80 are going to be extremely close in performance and features. I'm sure there's plenty of room for optimization and varying pipeline layouts, but who can say at this point which design will prove to be superior. Currently nVidia has the better track record in this area...but things change.
The main wild card I see is: did ATi opt for a 384bit memory bus (or perhaps 512bit)? If they didn't they might be in trouble. I don't really think 512bit is feasible at this point and I haven't heard anyone say R600 will have 384bit. Anyone hear anything about this?
Rangers
04-Nov-2006, 07:22
I'll just come out and say I hope ATI hits one out of the park with R600. Would certainly restore some faith in their engineering from me. As well they certainly need the good news. I'd even say if you like competition in this industry you should probably be rooting for R600 because if it flops..
The thing that's going to be hard for ATI is competing against the G80 refreshes again. Even if R600 would be a clear winner over G80, the margin might get smudged vs the refreshes, to the point where it's irrelevent, pretty much what happened with R520/80. ATI REALLY need to STOP being 3 months late..that three months is so crucial, all the high end adopters are always chomping at the bit, and they all want G80. You cant get that back.
Anyway, at the least it sounds like ATI will be back on more equal die size footing this time around, which will really help them even if say, the performance picture stays roughly the same as last gen, aka relative parity.
R600 will have about 40 million more transistors than G80. But if we take into account that secondary I/O chip, the overall count may be higher for G80.
Thats the first time that info is thrown around, where is it from.:huh:
Most of it from the Inq.....An example (http://www.beyond3d.com/forum/showpost.php?p=865881&postcount=2432) of why you might not want to use The Inq as a technical and especially a rumor source. Heck, even after the specs have been revealed by more knowledgable sources, don't put it past The Inq to still fumble the facts. You really have to know quite a bit before you're able to, well, safely interpret The Inq's rumors.
Anyway, given the relative paucity of consensus R600 specs, I'd wait until real G80 reviews before further guesstimating what R600 could/should be.
R600 will have about 40 million more transistors than G80
3 times more than Xenos?
An example (http://www.beyond3d.com/forum/showpost.php?p=865881&postcount=2432) of why you might not want to use The Inq as a technical and especially a rumor source. Heck, even after the specs have been revealed by more knowledgable sources, don't put it past The Inq to still fumble the facts. You really have to know quite a bit before you're able to, well, safely interpret The Inq's rumors.
Anyway, given the relative paucity of consensus R600 specs, I'd wait until real G80 reviews before further guesstimating what R600 could/should be.
I don't think it's a problem with their sources or their actual knowledge, it's simply they appear to believe anything and everything they read/hear. I wouldn't be surprised if there are people out there paid to be fake informants. It was nvidias interest that expectation for G80 wasn't sky high (had the real specs been leaked).
I'll sum it up with an example.
I once had an email conversation with one of the Inq editors, about 6 months ago.
He insisted that Vista would ship without the .Net framework installed, it would be an optional install on the disk. I asked for his reasons, and he sent a link to a ~2 year old article, warning that security changes in .Net 2 windows forms thread access (ie, creation thread can only change a form) may break backwards compatibility with 1.1 apps if 1.1 wasn't installed. From this he deduced that not only vista wouldn't include .Net at all, but .Net 'just isn't ready for primetime' (his words).
I'm not intending to derail the thread, just back up Pete :yes:
The only thing we can "hope" for is that R600 performs better than r520 vs. g70 when It comes out.
If R600 is anything like 520/580 and performs only marginally better than nV's then current offering, they will smack a GTX512/GX2 on the table and AMti are forced to yet again sit out another round in the GPU wars and the design teams will probably be forced into less spectacular projects.
R600 has to be what R300 was, I'm pretty sure they can't afford another year being the second best pick, especially if the sales in the other departments (cpu) are down as well management has to do something and by then it has to be a much better product than R400 ever was.
R600 has to be what R300 was, I'm pretty sure they can't afford another year being the second best pick, especially if the sales in the other departments (cpu) are down as well management has to do something and by then it has to be a much better product than R400 ever was.
ATI R500 Xenos - for Xbox360 was what R300 was.....
And we all hope R600 will be that way too, back in 2002 ATI had last chance to be worthy GPU company and ATI had success with R300. ATI simply can't afford screwup with R600.
wishiknew
05-Nov-2006, 00:00
I'm sure the r600 hardware will be in the ball park but its the software I'm worried about.
The r520's initial reviews had the bad 4xaa opengl memory controller mapping and adaptive aa didn't have the performance mode. The Avivo deinterlacing/acceleration was no go until 2 months later. H.264 got disabled later and re enabled later again. Multi sample adaptive aa mode only just now showed up using hidden registry keys. It took until cat 6.8 for major opengl increases which I assume come from the rewrite. Control panel still needs something like nvidia's lcd scaling options. Lets not even go into Crossfire.
Rangers
05-Nov-2006, 11:12
The only thing we can "hope" for is that R600 performs better than r520 vs. g70 when It comes out.
If R600 is anything like 520/580 and performs only marginally better than nV's then current offering, they will smack a GTX512/GX2 on the table and AMti are forced to yet again sit out another round in the GPU wars and the design teams will probably be forced into less spectacular projects.
R600 has to be what R300 was, I'm pretty sure they can't afford another year being the second best pick, especially if the sales in the other departments (cpu) are down as well management has to do something and by then it has to be a much better product than R400 ever was.
Well not really. There's no reason any of that "has" to happen at all. ATI will be in high end GPU's for at least another hundred years I'd guess.
I think if R600 is say, only 10-20% faster than G80, AMD will throw double the engineers into ATI to be good and sure to take the crown by a greater margin next time. And so on and so forth.
Where I see Nvidia begin to perhaps, focus more on there IGP business, and care less and less about the high end as time goes on. It's obvious to me Nvidia truly want to be in mobile markets most.
nicolasb
06-Nov-2006, 18:07
Well not really. There's no reason any of that "has" to happen at all. ATI will be in high end GPU's for at least another hundred years I'd guess.
I think if R600 is say, only 10-20% faster than G80, AMD will throw double the engineers into ATI to be good and sure to take the crown by a greater margin next time. And so on and so forth.Personally I doubt AMD has any particular interest in staying in the high-end graphics market at all. I think its motive in acquiring ATI was to gain expertise and IP in motherboard chipsets and integrated graphics. If AMD feels that it will make more money by taking all of the ATI personnel currently working in high-end graphics and redeploying them in other market segments (such as the "Fusion" project) that's exactly what it will do. My prediction is that, regardless of whether R600 soundly beats G80 or not, it will be ATI's last high-end graphics product (with the possible exception of an R620 or R680 minor revision).
Cuthalu
06-Nov-2006, 18:12
Personally I doubt AMD has any particular interest in staying in the high-end graphics market at all. I think its motive in acquiring ATI was to gain expertise and IP in motherboard chipsets and integrated graphics. If AMD feels that it will make more money by taking all of the ATI personnel currently working in high-end graphics and redeploying them in other market segments (such as the "Fusion" project) that's exactly what it will do. My prediction is that, regardless of whether R600 soundly beats G80 or not, it will be ATI's last high-end graphics product (with the possible exception of an R620 or R680 minor revision).
1) It might be economically better to be in the HE-business for ATi/AMD 2) ATi has already used resources for R700, so we're likely to see it in any case.
SugarCoat
06-Nov-2006, 18:23
Well not really. There's no reason any of that "has" to happen at all. ATI will be in high end GPU's for at least another hundred years I'd guess.
I think if R600 is say, only 10-20% faster than G80, AMD will throw double the engineers into ATI to be good and sure to take the crown by a greater margin next time. And so on and so forth.
Where I see Nvidia begin to perhaps, focus more on there IGP business, and care less and less about the high end as time goes on. It's obvious to me Nvidia truly want to be in mobile markets most.
What you're saying contradicts common sense. AMD cannot afford to just throw money into high end graphics development. I mean seriously, the best the thing anyone could honestly hope for is that AMD lets the graphics department operate off of their own budget. What ever they earn they can spend. Once you understand that, you'll see where the problem lies.
AMD is interested in getting more share and exposure in consumer desktops, they arent going to give a damn about crushing nVidia in the latest 3Dmark08 scores with the latest super expensive super elitist enthusiast card setups.
Geeforcer
06-Nov-2006, 18:50
Well not really. There's no reason any of that "has" to happen at all. ATI will be in high end GPU's for at least another hundred years I'd guess.
I think if R600 is say, only 10-20% faster than G80, AMD will throw double the engineers into ATI to be good and sure to take the crown by a greater margin next time. And so on and so forth.
Where I see Nvidia begin to perhaps, focus more on there IGP business, and care less and less about the high end as time goes on. It's obvious to me Nvidia truly want to be in mobile markets most.
This is almost the direct opposite of what I see happening. ATI is a low-margin, low-ROI business compared to AMD's core. AMD is far more likely to halve their resource allocation to it then double it, IMO.
Personally I doubt AMD has any particular interest in staying in the high-end graphics market at all. I think its motive in acquiring ATI was to gain expertise and IP in motherboard chipsets and integrated graphics. If AMD feels that it will make more money by taking all of the ATI personnel currently working in high-end graphics and redeploying them in other market segments (such as the "Fusion" project) that's exactly what it will do. My prediction is that, regardless of whether R600 soundly beats G80 or not, it will be ATI's last high-end graphics product (with the possible exception of an R620 or R680 minor revision).
The "next gen R600" has already been spotted on recent roadmaps for Q3'07, so a faster version of the R600 will definitely see the light. And don't forget AMDs recent presentation which also stated:
"Maintain and extend our leadership in discrete graphics"
Source: http://www.hardocp.com/news.html?news=MjIwODksLCxobmV3cywsLDE
That doesn't sound to me like their giving up on the high end in the long run.
Cuthalu
06-Nov-2006, 19:21
AMD is interested in getting more share and exposure in consumer desktops, they arent going to give a damn about crushing nVidia in the latest 3Dmark08 scores with the latest super expensive super elitist enthusiast card setups.
But that's common and good way to gain more share and especially exposure.
SugarCoat
06-Nov-2006, 19:31
But that's common and good way to gain more share and especially exposure.
they just dont have the money to toss into the fire like that.
Bouncing Zabaglione Bros.
06-Nov-2006, 19:59
Can you guys take this to one of the two "Fusion" threads and leave this one for R600 speculation?
Ok AMD, two days away from the G80 launch. Now would be a good time to start leaking.. :wink:
Bouncing Zabaglione Bros.
07-Nov-2006, 17:08
ATI better get it's finger out. Once again Nvidia's pulling a big crowd with their new tech, everyone is impressed, and by the time R600 arrives, unless it's even better by a significant margin, it will be a case of "meh, already seen that from Nvidia months ago."
Doesn't matter if R600 is as good as, or even a little bit better, Nvidia has already won the initial beachead into the "next-gen" battle, and ATI will be seen as followers rather than trailblazers.
Come on ATI, where are the spoilers? You know Nvidia will be doing it to you come your launch...
Galduta
07-Nov-2006, 23:02
One trustworthy store in Spain say that one ATI partner -that the usually makes products “of ice :wink: ” for Christmas- are saying to them that r600 is sent in January.
trinibwoy
08-Nov-2006, 00:00
January is fine by me!
Rangers
08-Nov-2006, 07:25
ATI better get it's finger out. Once again Nvidia's pulling a big crowd with their new tech, everyone is impressed, and by the time R600 arrives, unless it's even better by a significant margin, it will be a case of "meh, already seen that from Nvidia months ago."
Doesn't matter if R600 is as good as, or even a little bit better, Nvidia has already won the initial beachead into the "next-gen" battle, and ATI will be seen as followers rather than trailblazers.
Come on ATI, where are the spoilers? You know Nvidia will be doing it to you come your launch...
Ehh, I disagree, now that G80 has proven to be only 2X, if that, the 7900GTX. Kind of leaves the door open imo. I cant see any way ATI wont at least do that traditional 2X, so, R600 seems pretty safe, at the least, that it wont get left behind.
And we have already seen spoilers. 512bit, 160 GB/s BW..
Ailuros
08-Nov-2006, 07:29
Ehh, I disagree, now that G80 has proven to be only 2X, if that, the 7900GTX. Kind of leaves the door open imo. I cant see any way ATI wont at least do that traditional 2X, so, R600 seems pretty safe, at the least, that it wont get left behind.
To steal from one source is plagiarism, to steal from many research. Now keep that in mind before you jump to preliminary conclusions, because it sounds rather that the 8800GTS is roughly equal to a 7950GX2.
Chalnoth
08-Nov-2006, 08:08
Ehh, I disagree, now that G80 has proven to be only 2X, if that, the 7900GTX. Kind of leaves the door open imo. I cant see any way ATI wont at least do that traditional 2X, so, R600 seems pretty safe, at the least, that it wont get left behind.
And we have already seen spoilers. 512bit, 160 GB/s BW..
Except that these are still very early drivers on a radical change in architecture.
Bouncing Zabaglione Bros.
08-Nov-2006, 09:28
Ehh, I disagree, now that G80 has proven to be only 2X, if that, the 7900GTX. Kind of leaves the door open imo. I cant see any way ATI wont at least do that traditional 2X, so, R600 seems pretty safe, at the least, that it wont get left behind.
And we have already seen spoilers. 512bit, 160 GB/s BW..
But once again Nvidia get the early adopter and the mindshare for being a quarter early, just as they did with G70. A lot of people will view it as "Wow! G80 twice as fast as last generation", and then in a few months it will be "Meh, R600 ten percent faster than what we already have".
Just look around at the review and leaks. G80 is being compared to the previous generation cards and (of course) looking pretty impressive. In a few months, R600 is going to be compared to G80 of the same generation and won't have that wow factor. Sure, it's very superficial, but for most people, that's the way they will look at it.
Nvidia is even stealing ATI's "unified" thunder, though I'm not convinced you can call it unified if it has a fixed VS/PS ratio as reported.
Once again, this is ATI opening the door for Nvidia and inviting them through, with little if any response from ATI.
Ailuros
08-Nov-2006, 09:30
Nvidia is even stealing ATI's "unified" thunder, though I'm not convinced you can call it unified if it has a fixed VS/PS ratio as reported.
I wouldn't trust any BS that floats around.
Chalnoth
08-Nov-2006, 09:51
A fixed VS/PS ratio just makes no sense at all. Why take the majority of the cost of a unifed architecture, but almost none of the benefit?
Bouncing Zabaglione Bros.
08-Nov-2006, 09:53
I wouldn't trust any BS that floats around.
IMO, unless it's working like a unified architecture, then it's not one. It looks to me like Nvidia is adopting a marketing buzzword with a deliberately incorrect meaning in order to take away one of the big tech leads that ATI will be claiming. That's the kind of aggressive marketing that ATI should be combating and isn't.
Ailuros
08-Nov-2006, 09:54
IMO, unless it's working like a unified architecture, then it's not one. It looks to me like Nvidia is adopting a marketing buzzword with a deliberately incorrect meaning in order to take away one of the big tech leads that ATI will be claiming. That's the kind of aggressive marketing that ATI should be combating and isn't.
Then I guess you'll be unpleasantly surprised about how wrong you are.
Bouncing Zabaglione Bros.
08-Nov-2006, 09:55
A fixed VS/PS ratio just makes no sense at all. Why take the majority of the cost of a unifed architecture, but almost none of the benefit?
You're close to Nvidia's marketing machine, why don't you ask them and tell us? We'd all like to know how they can claim a unified architecture and yet say it's running in fixed ratios.
Geeforcer
08-Nov-2006, 09:55
IMO, unless it's working like a unified architecture, then it's not one. It looks to me like Nvidia is adopting a marketing buzzword with a deliberately incorrect meaning in order to take away one of the big tech leads that ATI will be claiming. That's the kind of aggressive marketing that ATI should be combating and isn't.
That "looks to you"... based on what???
Ailuros
08-Nov-2006, 09:56
You're close to Nvidia's marketing machine, why don't you ask them and tell us? We'd all like to know how they can claim a unified architecture and yet say it's running in fixed ratios.
For your information (and not that it should make a difference actually) he and I no longer are and that for quite some time now. That said you're still wrong.
Geeforcer
08-Nov-2006, 09:56
You're close to Nvidia's marketing machine, why don't you ask them and tell us? We'd all like to know how they can claim a unified architecture and yet say it's running in fixed ratios.
Where does it say that?
Bouncing Zabaglione Bros.
08-Nov-2006, 09:57
Then I guess you'll be unpleasantly surprised about how wrong you are.
Can you be a little more vague? :roll:
Are you saying that G80 is a unified architecture, that it's not running in fixed ratios, or that ATI will be combating Nvidia's aggressive marketing, or what?
Bouncing Zabaglione Bros.
08-Nov-2006, 09:58
Where does it say that?
Read through the G80 thread. It's in the leaks there somewhere. It's been quoted as being there "for compatability" and assumed to be fixed in future driver.
Ailuros
08-Nov-2006, 10:00
Can you be a little more vague? :roll:
Are you saying that G80 is a unified architecture, that it's not running in fixed ratios, or that ATI will be combating Nvidia's aggressive marketing, or what?
G80 is a USC and I have no reason to doubt that, especially considering the so far leaked results, indications, material whatever. Now you might want to see ghosts of weird conspiracy theories in every corner but there are people out there with experience that can analyze via software and see what what exactly is.
Ailuros
08-Nov-2006, 10:01
Read through the G80 thread. It's in the leaks there somewhere. It's been quoted as being there "for compatability" and assumed to be fixed in future driver.
Bears the question to you since you're the one making assumptions here: does that even make ANY sense?
Bouncing Zabaglione Bros.
08-Nov-2006, 10:05
Bears the question to you since you're the one making assumptions here: does that even make ANY sense?
No, and yet that's what's being said, and I certainly wouldn't put it past Nvidia to muddy the waters to their advantage. As you seem to be claiming definite knowledge, can you share some facts instead of just being vague?
Straight question: if I buy a G80 tomorrow, will it be a unified architecture, and will it load balance dynamically, or will it be set to run in fixed ratios?
Geeforcer
08-Nov-2006, 10:10
No, and yet that's what's being said...
It is?
What IS being said, exactly? The entire premise of your argument is based on the following sentence:
G80 is currently operating in the unified shader compatibility mode Logic plays a distributed implementation of vertex and pixel shader and geomery role Even in DirectX9.0C environment, the complexity of the order, intensive vertex / pixel geometry budget and the high load can play up to some extent reflects the unified shader power.
Soo... I'll repeat myself. Where does it say anything about ratios being fixed?
Ailuros
08-Nov-2006, 10:13
Your question builds an absolute oxymoron; that's the one point.
The other point being that even if I would have any classified info I wouldn't be as naive to run around and share it.
Bouncing Zabaglione Bros.
08-Nov-2006, 10:16
Well you keep saying I'm wrong, yet you won't correct me, so I can only surmise that you don't really know either, despite you working hard to give the impression that you do.
Geeforcer
08-Nov-2006, 10:20
Well you keep saying I'm wrong, yet you won't correct me, so I can only surmise that you don't really know either, despite you working hard to give the impression that you do.
What would he correct? Right now, there is nothing. Maybe if there was any evidence to support what you have said this discussion would go somewhere.
Ailuros
08-Nov-2006, 10:24
Well you keep saying I'm wrong, yet you won't correct me, so I can only surmise that you don't really know either, despite you working hard to give the impression that you do.
I asked a very simple and straightforward question to which I haven't received an answer yet. What is there to correct exactly? It might be a nice way to paddle yourself out of it and try to discredit what I am trying to say, but at least bother to look at the so far leaked results.
Bouncing Zabaglione Bros.
08-Nov-2006, 10:27
I asked a very simple and straightforward question to which I haven't received an answer yet. What is there to correct exactly? It might be a nice way to paddle yourself out of it and try to discredit what I am trying to say, but at least bother to look at the so far leaked results.
DX10 performance can not be tested yet and in DX9 applications the unified architecture works as a fixed function architecture (dedicated pixelshaders and vertex shaders).
Here you go. (http://www.beyond3d.com/forum/showpost.php?p=864210&postcount=2257) This is where the info first surfaced. CJ has been spot on, to the extent that B3D mods recently removed his specification info from yesterday's G80 thread because they don't want to breach NDA.
Now will you answer my question? Or at least tell me that you don't really know? Because it really makes no sense to me to have a unified architecture and then make it perform as if it's a fixed architecture.
IMO, unless it's working like a unified architecture, then it's not one. It looks to me like Nvidia is adopting a marketing buzzword with a deliberately incorrect meaning in order to take away one of the big tech leads that ATI will be claiming. That's the kind of aggressive marketing that ATI should be combating and isn't.
it seems to me that you making up stuff, why you just don't wait a few hours so that you (and everyone else) can have a clearer picture?
Bouncing Zabaglione Bros.
08-Nov-2006, 10:34
it seems to me that you making up stuff, why you just don't wait a few hours so that you (and everyone else) can have a clearer picture?
That's ironic given that you wrote this (http://www.beyond3d.com/forum/showpost.php?p=864215&postcount=2259).
This sound as "we still don't get difference between having a unified shading sw model and a unified hw shading architecture", therefore it's probably BS :)
Ailuros
08-Nov-2006, 10:40
Here you go. (http://www.beyond3d.com/forum/showpost.php?p=864210&postcount=2257) This is where the info first surfaced. CJ has been spot on, to the extent that B3D mods recently removed his specification info from yesterday's G80 thread because they don't want to breach NDA.
Spot on on what exactly? That the GPU is "CPU limited" or the senseless DX10 reference?
Now will you answer my question? Or at least tell me that you don't really know? Because it really makes no sense to me to have a unified architecture and then make it perform as if it's a fixed architecture.
Not only doesn't it make sense I can't even imagine a way that could get you something like that done.
In the meantime have a look at the triangle rates in the Archmark results and start wondering if there's anything worthwhile noticing:
http://www.beyond3d.com/forum/showpost.php?p=866340&postcount=39
Now that silly debate whether I do or don't know anything is kindergarten material at best. If you can stick to the point with something that makes sense apart from some wild claims from someone that translated something which obviously hasn't much of a clue of what he was looking at isn't really material to debate about.
LOL! that 'we' is not referred to nvidia but to the guys writing that BS, shoud I include you in the list of those that still does not grasp that thing as well?
Even if DX9 has not a unified shading model IT DOES NOT MEAN that a GPU employing a unified architecture should work without dynamic balancing while it's running DX9 apps.
Hope it's clear now :)
Bouncing Zabaglione Bros.
08-Nov-2006, 10:55
Not only doesn't it make sense I can't even imagine a way that could get you something like that done.
Load balancing and unifed don't necessarily go hand in hand (http://www.beyond3d.com/forum/showpost.php?p=864268&postcount=2270).
Now that silly debate whether I do or don't know anything is kindergarten material at best. If you can stick to the point with something that makes sense apart from some wild claims from someone that translated something which obviously hasn't much of a clue of what he was looking at isn't really material to debate about.
Well when someone says I'm talking nonsense, I like to be educated by finding out what is correct. That is pretty much what B3D is about. You keep telling me I've got the wrong answer, but you won't tell me the right answer, and yet you seem to be basing all your info on stuff you've read on the forums about G80. You won't confirm that you actually know facts, that you have had a G80 in your hand and run tests, that you have an NDA, etc, you just keep implying that you know better without backing it up.
If you can give a statement of fact that is more than conjecture on your own part, then please do so. Even if it's just to say you've had a G80 and you know what they are. I don't think that's a silly kindergarten thing to ask for.
Ailuros
08-Nov-2006, 10:59
If you can give a statement of fact that is more than conjecture on your own part, then please do so. Even if it's just to say you've had a G80 and you know what they are. I don't think that's a silly kindergarten thing to ask for.
Did you even bother to open the link in my former post? Notice the 8light geometry score? Anything weird compared to other GPUs?
Bouncing Zabaglione Bros.
08-Nov-2006, 11:03
Did you even bother to open the link in my former post? Notice the 8light geometry score? Anything weird compared to other GPUs?
Why don't you explain it to me, as the link you gave doesn't show any other GPU results? Imagine I'm an idiot.
So is that what you're basing your assertion on? Or have you had a G80 yourself?
I wouldn't trust any BS that floats around.And there's enough of it floating around to fill the grand canyon twice over.
Ailuros
08-Nov-2006, 11:17
Here are comparable results:
http://www.beyond3d.com/forum/showpost.php?p=866457&postcount=54
And here some food for thought:
http://www.beyond3d.com/forum/showpost.php?p=866524&postcount=61
Bouncing Zabaglione Bros.
08-Nov-2006, 11:27
Here are comparable results:
http://www.beyond3d.com/forum/showpost.php?p=866457&postcount=54
And here some food for thought:
http://www.beyond3d.com/forum/showpost.php?p=866524&postcount=61
Hmm, so would this be due to the G80 having only unified PS/VS? In this case the even with all the shaders dedicated to VS, it gets choked on the triangle setup?
So, would a fully unified and load balanced architecture as we're assuming R600 to be (see that - back on topic?), be able to deal with this by dynamically dedicating more of it's transistor pool to triangle setup rather than having some of the VS/PS pool go idle while waiting for triangle setup? Or does fully unified really just mean VS/PS and not other functions like triangle setup?
Ailuros
08-Nov-2006, 11:33
And there's enough of it floating around to fill the grand canyon twice over.
PC Online slides containing detailed block diagrams are still available (albeit they've pulled the front page only temporarily). I could easily copy/paste them and upload them. The first thing that speaks against it is that I don't want to embarass the B3D folks and the second being that he'll simply disregard it because it comes from NV. In the meantime some hearsay from someone that claims G80 to be "CPU limited" is more credible than the IHVs claims itself.
No wonder I feel more than often diagonally parked in a parallel universe; especially if the Grand Canyon seems to swap over way more than twice.
Bouncing Zabaglione Bros.
08-Nov-2006, 11:38
There'd be no point in the existence of B3D if all we had to do was take the manufacturer's marketing material as the be-all and end-all. Especially given how "flexible" with the truth they've been in the past.
Ailuros
08-Nov-2006, 11:45
Hmm, so would this be due to the G80 having only unified PS/VS? In this case the even with all the shaders dedicated to VS, it gets choked on the triangle setup?
Unified GS/VS/PS. Yes and Chal's explanation makes perfect sense to me.
So, would a fully unified and load balanced architecture as we're assuming R600 to be (see that - back on topic?), be able to deal with this by dynamically dedicating more of it's transistor pool to triangle setup rather than having some of the VS/PS pool go idle while waiting for triangle setup? Or does fully unified really just mean VS/PS and not other functions like triangle setup?
IHVs estimate the needed performance characteristics for a given timeframe and will always set an upper threshold on the triangle setup as there of course is on Xenos. More would mean redundant hardware for any given timeframe. In theory Xenos someone could easily say that 48 ALUs * 500MHz / 4= 6000 MVertices/s or an analogue rate for triangles. The reason why it'll never reach that theoretical rate is that there is in fact a set upper threshold for the triangle setup which isn't exactly small for such a GPU to be honest.
Ironically Bob is an engineer that asked that question above; I rather suspect he asked it on purpose.
trinibwoy
08-Nov-2006, 12:00
So possible shortcomings of G80 now count as R600 rumours? :lol:
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.