Huddy says "R600"

[Chalnoth]

Well, sure, it's a burden on IHV's, but it's not like they won't have seen it coming. They'll probably both have tried to keep the architectures similar enough so that they can basically use their current DX9 drivers tweaked slightly for the new architecture.

[pharma]

I read most of this thread but did not see this mentioned ... R600 will meet Vista requirements for DX10, but will it also be able to run on other OS such as XP or Linux using DX9? :?: Or not?

Pharma

[karlotta]

I read most of this thread but did not see this mentioned ... R600 will meet Vista requirements for DX10, but will it also be able to run on other OS such as XP or Linux using DX9? :?: Or not?

Pharmaof coures it will. I think it is said by huddy, or ATI that the R600 will be the fastest dx9 card ever. And of course a D10 card.

[pharma]

of coures it will. I think it is said by huddy, or ATI that the R600 will be the fastest dx9 card ever. And of course a D10 card.

Thanks, I thought so but some posts in this thread confused me regarding XP lacking DX10 and the need to upgrade to Vista. But I guess that's only if you want to have the full effect of the Unified Architecture.

Pharma

[tEd]

of coures it will. I think it is said by huddy, or ATI that the R600 will be the fastest dx9 card ever. And of course a D10 card.

d3d9 also exist on vista so that's not a 100% confirmation that there will be xp drivers

[Tim Murray]

d3d9 also exist on vista so that's not a 100% confirmation that there will be xp drivers
except both G80 and R600 will be coming quite a while before Vista.

[SugarCoat]

no offence but some people are being dumb. ATi and Nvidia will continue to support all previous versions of Direct X, especially for WinXP, for quite some time.

[INKster]

no offence but some people are being dumb. ATi and Nvidia will continue to support all previous versions of Direct X, especially for WinXP, for quite some time.

Well, they only really need to have support for DX9 and DX10, since all the others are obsolete.
Besides, DX9 is backwards compatible with earlier versions, right ?


I don't doubt they'll continue to support WinXP/2000 for a few years also, but, since WVDDM and WDM driver models are so different, it may not be such a long term commitment as with the simultaneous W9x/W2K/WXP DirectX one.

It all comes down to the market adoption rate of both Vista/DX10 software and DX10 ready hardware by the time they'll have to make that call, business-wise.

[Moloch]

Well, they only really need to have support for DX9 and DX10, since all the others are obsolete.
Besides, DX9 is backwards compatible with earlier versions, right ?


I don't doubt they'll continue to support WinXP/2000 for a few years also, but, since WVDDM and WDM driver models are so different, it may not be such a long term commitment as with the simultaneous W9x/W2K/WXP DirectX one.

It all comes down to the market adoption rate of both Vista/DX10 software and DX10 ready hardware by the time they'll have to make that call, business-wise.
Obsolete or not, some people still play DX8 games, even DX7 :!:

[Chalnoth]

Obsolete or not, some people still play DX8 games, even DX7 :!:
Right, but the driver interface is the same for both. Windows basically handles the backwards compatibility (though I'm sure each IHV has to do compatibility testing, obviously).

[_xxx_]

Obsolete or not, some people still play DX8 games, even DX7 :!:

I even still play some DOS games...

[Dave Baumann]

General note: Theres no technical reasons why DX10 boards will not operate under WinXP with DX9 being the maximum capability (irrespective of the architecture). I think IHV's will have to maintain the following different software levels: DX9 board - WinXP DX9, Longhorn Driver model DX9; DX10 board - WinXP DX9, Longhorn Driver Model DX10.

But I guess that's only if you want to have the full effect of the Unified Architecture.
Unified or not is not a function of the API environment is operating under, but an implementation detail of the architecture. You get the "full effect" of the unified architecture itself whether running under DX9 or DX10; you will miss out on some of the full capabilities but that would be the same as being unified or not.

except both G80 and R600 will be coming quite a while before Vista.
Bear in mind that there are versions of Vista coming in 2006 (the non-home versions) - the business/OEM market is unlikely to be putting R600/G80 in their boxes, but inevitably those versions of Vista will end up on sale by one method or another.

[MistaPi]

Bear in mind that there are versions of Vista coming in 2006 (the non-home versions) - the business/OEM market is unlikely to be putting R600/G80 in their boxes, but inevitably those versions of Vista will end up on sale by one method or another.
Are you talking about the hole Vista product range? I thought only the server editions would be released this year.

[INKster]

Are you talking about the hole Vista product range? I thought only the server editions would be released this year.

No, the server versions of Longhorn/Vista will come out roughly 1 year later, late 2007/early 2008 (hopefully).

[SugarCoat]

http://www.xbitlabs.com/news/video/display/20060621042925.html


Sounds like new mainstream parts in Q1 07.


Oh and here you go Geo, week late ;P

http://www.theinq.net/?article=32546

[Geo]

Thanks, I think that's the first indication we've had on a target date for the rest of the R6xx family. :grin:

[Demirug]

General note: Theres no technical reasons why DX10 boards will not operate under WinXP with DX9 being the maximum capability (irrespective of the architecture). I think IHV's will have to maintain the following different software levels: DX9 board - WinXP DX9, Longhorn Driver model DX9; DX10 board - WinXP DX9, Longhorn Driver Model DX10.

As far as I know you still have to write the D3D9 driver part for Vista even if you have a D3D10 board.

[Chalnoth]

As far as I know you still have to write the D3D9 driver part for Vista even if you have a D3D10 board.
I'm not sure you need a different driver for Vista's D3D9 interface than you do for their D3D10 interface.

[Rys]

I think you definitely do.

[Demirug]

I'm not sure you need a different driver for Vista's D3D9 interface than you do for their D3D10 interface.

The functions the driver has to provide for D3D9 and D3D10 are different.
If you additional compare the OpenAdapter (D3D9) and OpenAdapterD3D10 function you will see two different descriptions for the interface member in the create structure

OpenAdapter:
Interface
[in] The Direct3D/DirectDraw interface version (for example, 7, 8, or 9) that opens the graphics adapter.

OpenAdapterD3D10:
Interface
[in] The Direct3D or DirectDraw interface version. The high 16 bits store the major release number (such as 10, 11, and so on); the low 16 bits store the minor release number (such as 0, 1, 2, and so on).

Sound for me like any runtime below D3D10 use the D3D9 driver interface.

[Chalnoth]

Well, I guess I don't know that much about Windows drivers, but is that a function that the IHV's have to supply in their driver to Direct3D, or is that a function that programmers use to query the driver about capabilities?

[Demirug]

Well, I guess I don't know that much about Windows drivers, but is that a function that the IHV's have to supply in their driver to Direct3D, or is that a function that programmers use to query the driver about capabilities?

The IHV have to supply these functions in the user mode driver. There are many more. As a developer you normally don’t call these functions direct. You call functions in the runtime and the runtime calls this driver functions. It is possible to talk direct to the driver (I have done this) but you will lose compatibility every time Microsoft change the driver interface.

[Chalnoth]

The IHV have to supply these functions in the user mode driver. There are many more. As a developer you normally don’t call these functions direct. You call functions in the runtime and the runtime calls this driver functions. It is possible to talk direct to the driver (I have done this) but you will lose compatibility every time Microsoft change the driver interface.
Ah, I see. Well, I find it rather likely that there won't be two separate drivers for Vista, but rather just different function calls made to the same slightly abstracted interface for the core driver.

Obviously IHV's will need to support the DX9 interface under Vista, otherwise no current games would run on new hardware in Vista, which would be a perfect way to lose sales.

[Demirug]

Ah, I see. Well, I find it rather likely that there won't be two separate drivers for Vista, but rather just different function calls made to the same slightly abstracted interface for the core driver.

For sure both interfaces will be integrated in one driver (and the OpenGL part too). I am even very sure that both interfaces will be in same DLL as there should be many code that are need independent for the interface.

The kernel mode driver is already more interface independent.

[SugarCoat]

Ah, I see. Well, I find it rather likely that there won't be two separate drivers for Vista, but rather just different function calls made to the same slightly abstracted interface for the core driver.

Obviously IHV's will need to support the DX9 interface under Vista, otherwise no current games would run on new hardware in Vista, which would be a perfect way to lose sales.


I think you have it backwards, if current beta Vista drivers are missing anything its DX10 functionality. If Nvidia and ATI driver teams have been working on anything it is infact the DX9 driver for Vista. I have even gotten DX7 games to run on RC2 without issue. And even if it requires two seperate drivers, the end user shouldnt really notice since the Vista driver will hopefully be 1 package.

[Chalnoth]

I think you have it backwards, if current beta Vista drivers are missing anything its DX10 functionality. If Nvidia and ATI driver teams have been working on anything it is infact the DX9 driver for Vista. I have even gotten DX7 games to run on RC2 without issue. And even if it requires two seperate drivers, the end user shouldnt really notice since the Vista driver will hopefully be 1 package.
Well, the current Vista beta drivers are for DX9 hardware, aren't they? :)

But either way, there should be no worries about drive incompatibility with DX9 for DX10 hardware, either in Vista or in XP or previous.

[SugarCoat]

Well, the current Vista beta drivers are for DX9 hardware, aren't they? :)

But either way, there should be no worries about drive incompatibility with DX9 for DX10 hardware, either in Vista or in XP or previous.


Yes they are. But your comment was about them having to get DX9 working on Vista or sales would plummit, so i was trying to assure you its working quite well. From DX9 and all the way down.

[Chalnoth]

Hey, I'm not the one that was making any claims about lack of backwards-compatibility on DX10 hardware, either in Vista or XP.

[pharma]

RS600 to lack DirectX9c support?


Since the RS600 uses a Radeon X700 derived graphics core it lacks support for DirectX 9c and Shader Model 3.0. Roadmaps show ATI will skip DirectX 9c support for its integrated graphics core and instead jump to DirectX 10.


http://dailytech.com/article.aspx?newsid=2958

[Dave Baumann]

DirectX9c is not necessarily related to SM3.0. In fact, if its based on X700, as it states, then it will utilise functionality that DX9.0c makes available as this brought the PS2.b profile. RS600 has long since said the be X700 based, so SM2 capabilities is a given, but seeing that its primary 3D task will likely be running the Vista interface no higher shader model is required (in fact, by just implementing SM2 hardware you can probably increase the overall performance by including more more pipelines, in these cost consious applications, as opposed to using a bunch of transistors for SM3.0 capabilities - certainly with ATI's SM3.0 design choices).

[Tahir2]

I would agree the extra transistors required for SM3.0 (esp. ATI's "version" of it) would be better spent on adding pipelines in an integrated solution.

[trumphsiao]

some AIB partners have been received G80 samples . :grin:

[Geo]

some AIB partners have been received some G80 samples . :grin:

You have something to point at, or was there a note pushed under your door over night? :smile:

[SugarCoat]

Phase 1 is complete. GPU taped out reports.
Phase 2 is in process. Waiting on leaked Die shots.

[Geo]

Oh, that's not phase II. That's several steps down the road. One of these days I'm really going to do that consolidated milestones on the internet road to a new GPU-gen release chart.

AIB's getting samples (if true) is a good sign. Starting to see evidence that ISVs have as well would be another one. . .

[trumphsiao]

Oh, that's not phase II. That's several steps down the road. One of these days I'm really going to do that consolidated milestones on the internet road to a new GPU-gen release chart.

AIB's getting samples (if true) is a good sign. Starting to see evidence that ISVs have as well would be another one. . .


I hope this time history wont repeat itself. ATI should not take the premature and measly capacity process(TSMC 65nm Process ) instead of 90/80nm Process with enormous capacity . I dont want to see repetition of last year's tragic events again.

[Subtlesnake]

Last year's events occurred because there was a bug in some third party libraries.

According to ATI the transition to 90 nm was the smoothest they'd ever made, and when the X1800 was finally ready, ATI was able to get excellent yields.

[IbaneZ]

some AIB partners have been received G80 samples . :grin:

:shock:

Link/source? :wink:

[no-X]

Fuad says: ATI R600 to get a new cooler

http://www.theinq.com/default.aspx?article=32712


already mentioned: ATI R600 will run DX9 faster

http://www.theinq.com/default.aspx?article=32713

[_xxx_]

:shock:

Link/source? :wink:

"A reliable source he can't disclose", I suppose :razz:

[pjbliverpool]

Fuad says: ATI R600 will run DX9 faster

http://www.theinq.com/default.aspx?article=32713

Another golden nugget of information from Fuad. What would we do without his insider knowledge!

[nAo]

Another golden nugget of information from Fuad. What would we do without his insider knowledge!
well..ATI is getting very good at using him..

[trumphsiao]

Another golden nugget of information from Fuad. What would we do without his insider knowledge!


R600 is on TSMC 65nm Process. failing that .ATI will have to procrastinate all R6XX/RV6XX SKUs .

[SugarCoat]

R600 is on TSMC 65nm Process. failing that .ATI will have to procrastinate all R6XX/RV6XX SKUs .


A question was asked as to what process their next generation architecture would be based on, and Dave Orton pointed out that the 80nm process comes in a number of flavours, including a cost reduction option, which is currently in the process of being adopted for ATI’s low end parts, and also an 80nm HS process, to be used on more expensive higher end solutions. Dave then went on to say that none of the R6xx generation is likely to be 90nm based, instead split between 80nm and 65nm processes.

http://www.beyond3d.com/forum/showthread.php?t=29612


If the flagship chips are as big and fast as they say and the chips are indeed 65nm then 65nm fab at TSMC will problably be dedicated to only those flagship chips due to the relatively small percent 65nm fab will make up that early in the year. Essentially 65nm fab should only be dedicated to priority parts where they can benefit most. Cheap low end and problably some of the mid-end parts will problably be 80nm. I'm still not use to the idea of the R600 series flagship being 65nm so early though. Will be quite a shocker for me once confirmed.

[Chalnoth]

Splitting of the R6xx generation between 80nm and 65nm does not imply that any of the intial parts in the generation will be 65nm. I think we can reasonably expect their initial R6xx high-end offering to be 80nm, with a 65nm part to come later.

[SugarCoat]

depends how late the part is in my opinion. The fact that the date is jumping from early Q4 2006 to Q1 2007 seems strange. Hope they arent having issues again. Nvidia may cash in on early Vista ready OEM sales if they can hit september area.

[Razor1]

I don't think nV will be ready with the g80 around september. As it is right now, The r580+ will have to be clocked pretty high to take on the 7950. Also if that 500+ mill trany count is correct, the g80 has to be on .8, which doesn't seem that feesable in September.

[SugarCoat]

http://www.theinquirer.net/default.aspx?article=33093

[Geo]

Damn him! Is he trying to suggest that R600 is 17% smaller than R580? Or just that 90nm is 17% smaller than 80nm?

[_xxx_]

I think you should block their IP-range from accessing B3D so they stop stealing here ;)

[Sunrise]

Damn him! Is he trying to suggest that R600 is 17% smaller than R580? Or just that 90nm is 17% smaller than 80nm?
That´s simply a 90nm -> 80nm comparison, which he also did wrong, because it should be "up to 19 percent" and not 17 percent.

[rwolf]

That´s simply a 90nm -> 80nm comparison, which he also did wrong, because it should be "up to 19 percent" and not 17 percent.

That is what I thought too.

[CosmoKramer]

That is what I thought too.

Maybe he takes the chip padding into account.

[Geo]

Moved the G80 posts to the G80 thread: http://www.beyond3d.com/forum/showthread.php?t=28551&page=17

[LeStoffer]

No tidbits about the R600 in a long time. :sad:

Since it's on 80nm and ATI had to delay the Gemini-launch (possibily because of ongoing tweaks to the new process) the R600 got in problems as well. So here's me guess: The much more complex R600 had to have at least one more spin than the Gemini chips, so we can at best expect X2000 to ship 6 weeks after the Gemini introduction. That would point to the last week of November at the very earliest. *crosses fingers*

[trumphsiao]

No tidbits about the R600 in a long time. :sad:

Since it's on 80nm and ATI had to delay the Gemini-launch (possibily because of ongoing tweaks to the new process) the R600 got in problems as well. So here's me guess: The much more complex R600 had to have at least one more spin than the Gemini chips, so we can at best expect X2000 to ship 6 weeks after the Gemini introduction. That would point to the last week of November at the very earliest. *crosses fingers*



R600 will be a little further delayed than we expected.
600M ,80nm, 370mm^2~390mm^2 is the rumor configuration among AIBs.

[satein]

R600 will be a little further delayed than we expected.
600M ,80nm, 370mm^2~390mm^2 is the rumor configuration among AIBs.

Do you know what would be reasonable reasons those are responsible for this delay? May it concern anything related to the merger between AMD/ATi recently? Hopefully, ATi can keep up their scheduling either sooner or later :roll:

[RejZoR]

Well NVIDIA is certanly in a very good position at the moment. They'll have the first DX10 part sent to market as being the fastest DX9 card to date and if they've done it well, also the fastest card for DX10 when Vista ships. ATI on the other hand can only catch a train for DX10 where it can take the "crown" from NVIDIA. AMD merger stuff isn't really helping them (the ATi), but is certanly helping NVIDIA guys...

[LeStoffer]

R600 will be a little further delayed than we expected.

December or January?


600M ,80nm, 370mm^2~390mm^2 is the rumor configuration among AIBs.

600M sounds doable, but 370mm^2~390mm^2 isn't much larger than the R580 even given the move to 80nm?

[neliz]

AMD merger stuff isn't really helping them (the ATi), but is certanly helping NVIDIA guys...

What has the merger got to do with a possible delay of the R600? This time last year we were all expecting G80 to be out last May/June and R600 to follow suit in October, so far both companies do not need to release the card, sans lack of Vista.

[Dave Baumann]

This time last year we were all expecting G80 to be out last May/June...

Kind of ironic that, isn't it...

http://www.beyond3d.com/forum/showthread.php?t=26172

[INKster]

I think Microsoft ended up delaying the timetable of both ATI and Nvidia because of the Windows Vista "train-wreck".

[neliz]

I think Microsoft ended up delaying the timetable of both ATI and Nvidia because of the Windows Vista "train-wreck".

What about breathing space?

Vista delay maybe bad for PC sales in general, but I'm sure everyone at nV/Ati is happy to not have a tight noose around they're neck with regards to dx10

[trumphsiao]

Kind of ironic that, isn't it...

http://www.beyond3d.com/forum/showthread.php?t=26172


Dear Dave Baumann. tho previous Lord of B3D. Can you steer us a clear way out about Dual-Core R600 is on the way or just a falsification ?.:grin:

[INKster]

What about breathing space?

Vista delay maybe bad for PC sales in general, but I'm sure everyone at nV/Ati is happy to not have a tight noose around they're neck with regards to dx10

But, since DX10 is substancially different from DX9, no one would expect DX10-optimized software anytime soon.
This would allow them both to optimize the hardware mostly for DX9, buying them the extra time anyway.
When DX9 first showed up in 2002, even true DX8 games were still few and far between.

[trumphsiao]

But, since DX10 is substancially different from DX9, no one would expect DX10-optimized software anytime soon.
This would allow them both to optimize the hardware mostly for DX9, buying them the extra time anyway.
When DX9 first showed up in 2002, even true DX8 games were still few and far between.


I sincerely concur your comment . The hydra befuddle me is How much game dovetailed advancement and entertainment altogether are bricolages when G80 and R600 are on the shelves ?

[trinibwoy]

I sincerely concur your comment . The hydra befuddle me is How much game dovetailed advancement and entertainment altogether are bricolages when G80 and R600 are on the shelves ?

Your hydra befuddles me as well :cry:

[neliz]

But, since DX10 is substancially different from DX9, no one would expect DX10-optimized software anytime soon.
This would allow them both to optimize the hardware mostly for DX9, buying them the extra time anyway.

Why would hardware need to be optimized for DX9? , G80 is an evolution, G71 on steroids. R600 is a revolution, which also provides enough horsepower to beat current gen cards.



When DX9 first showed up in 2002, even true DX8 games were still few and far between.

When DX5 appeared, hardly any game needed it. the R300 was the DX9 slap in the face that nobody expected, but seriously the industry is evolving. I still think Halo2 for Vista will be DX10 only, but seeing the broad support dx10 is getting from the industry I hardly think a "hey, but it used to be like this" stance is a good one to take.

[INKster]

Why would hardware need to be optimized for DX9? , G80 is an evolution, G71 on steroids. R600 is a revolution, which also provides enough horsepower to beat current gen cards.

Perhaps because no DX10 games exist yet ?
For the next year or so, 90%+ of all games will still be DX9, most likely.
And it's not just the "new API" stigma.
Games are developed for the greatest installed base in order to maximize sales, and that base right now is DX9.


When DX5 appeared, hardly any game needed it. the R300 was the DX9 slap in the face that nobody expected, but seriously the industry is evolving. I still think Halo2 for Vista will be DX10 only, but seeing the broad support dx10 is getting from the industry I hardly think a "hey, but it used to be like this" stance is a good one to take.

Back when DX5 appeared, most 3D graphics processors relied on proprietary API's (Glide, PowerSGL, etc) and OpenGL.
It was not until DX7 that Microsoft's API began to loose the "inferior" reputation in that regard.

[neliz]

Perhaps because no DX10 games exist yet ?
For the next year or so, 90%+ of all games will still be DX9, most likely.
Did the egg beat the chicken to it?

No dx10 cards = no dx10 games
there can't be a dx10 game without a dx10 card eh?
Hell, if Vista launches, even S3 will have a DX10 card





Back when DX5 appeared, most 3D graphics processors relied on proprietary API's (Glide, PowerSGL, etc) and OpenGL.
It was not until DX7 that Microsoft's API began to loose the "inferior" reputation in that regard.
Right, There will be some breathing space for DX9 cards during the first year or so, but I'm sure there are advantages to going DX10 that are incentives for developers to go dx10 sooner rather than later.

[trinibwoy]

Why would hardware need to be optimized for DX9? , G80 is an evolution, G71 on steroids. R600 is a revolution, which also provides enough horsepower to beat current gen cards.

You may be right - the general opinion seems that the only way to revolutionize is to go unified. But hopefully we will be pleasantly surprised by G80 in terms of architectural sophistication.

[trumphsiao]

Your hydra befuddles me as well :cry:



sorry for my misused word and my poor english.
http://www.thefreedictionary.com/hydra

[karlotta]

Did the egg beat the chicken to it?

No dx10 cards = no dx10 games
there can't be a dx10 game without a dx10 card eh?
Hell, if Vista launches, even S3 will have a DX10 card





Right, There will be some breathing space for DX9 cards during the first year or so, but I'm sure there are advantages to going DX10 that are incentives for developers to go dx10 sooner rather than later. Yes a sticker on the box"DX10", Big dev house's can use CPU/SW to run beta DX10, before they have a card?

[neliz]

But hopefully we will be pleasantly surprised by G80 in terms of architectural sophistication.

well, it will need new logic and a lot of it for dx10, but I can't get rid of the feeling that nV took their time with nv4x and whatever they had ready for nv5x and experiment, things that are in g7x which gives it some benefits.

[Sunrise]

Your hydra befuddles me as well :cry:
:lol: (I laughed so hard i almost broke out in tears.)

[trumphsiao]

:lol: (I laughed so hard i almost broke out in tears.)


give me a tissue paper asap.
:wink:

[INKster]

Did the egg beat the chicken to it?

No dx10 cards = no dx10 games
there can't be a dx10 game without a dx10 card eh?
Hell, if Vista launches, even S3 will have a DX10 card

Well, you might get a "DX10 game" as advertized by the publisher, but with only minor differences in general look and feel, compared to DX9.
It's one thing to simply "port" to DirectX 10, and add a couple of new eye-candy effects to the game and say it's there only because DX10 enables it (even though it might have been accomplished in DX9, given the time/effort/motivation).

It's another altogether when you have to build your engine thinking about DX10, implementing true innovation right from the start.



Right, There will be some breathing space for DX9 cards during the first year or so, but I'm sure there are advantages to going DX10 that are incentives for developers to go dx10 sooner rather than later.

As this industry has proved in recent years, devs just keep loosing power to the bean counters.
If they say:
"game x must come out on date y, or we'll go overbudget", or "do what you have to do to get something on it -marketing purposes, mostly- so we can say DX10 too in the adverts", devs have to scale down on new projects, or they'll need more time.

Very few devs can afford the luxury of saying "we'll ship only when it's ready", at least not out loud...

[trinibwoy]

well, it will need new logic and a lot of it for dx10, but I can't get rid of the feeling that nV took their time with nv4x and whatever they had ready for nv5x and experiment, things that are in g7x which gives it some benefits.

Well they took their time with nv4x and everything since then has just been relatively minor refinements of the same architecture. Even transparency-AA was already in NV40. What have they been laboring on since then? Of course there are things in G7x which gives it some benefits just like there are things in R5x which benefit R600.

Maybe I'm off but it's seems a bit unlikely that all Nvidia could come up with was "G71 on steroids" while ATi was slugging away at R600. Especially considering that we've already seen much of R600's lineage in Xenos and R5x - we haven't seen anything "new" from Nvidia in over two years. It's what we haven't seen that makes me optimistic about G80.

[trinibwoy]

:lol: (I laughed so hard i almost broke out in tears.)

Glad to be of service :lol:

[Chalnoth]

Well they took their time with nv4x and everything since then has just been relatively minor refinements of the same architecture.
Well, right. Architectural cycles are lengthening. It's only been just barely over two years. We might possibly expect the next cycle to last three.

[DmitryKo]

Well, you might get a "DX10 game" as advertized by the publisher, but with only minor differences in general look and feel, compared to DX9 To be precise, there's no such thing as DirectX 10. There is Direct3D10 (which is the name that have always been used at the API level) and anything else in the DirectX runtime is being deprecated. So basically the only official sticker available is "Designed for Windows Vista", though I can see how developers and vendors will still be touting "DirectX 10" or even "DirectX Graphics 10" etc.

[swaaye]

Well consider Guild Wars was using a DX8 renderer until recently.....

I think the requirement to switch to a new OS is going to really curtail DX10's adoption. It's a bit overly convenient for MS IMO, and ridiculous too I think.

[Razor1]

I think the requirement to switch to a new OS is going to really curtail DX10's adoption. It's a bit overly convenient for MS IMO, and ridiculous too I think.

Thats my feeling also, plus the Dx10 sdk has been out for a few months already publicly with the dx9 sdk, and a few months before that for developers.

[ivzk]

Well NVIDIA is certanly in a very good position at the moment. They'll have the first DX10 part sent to market as being the fastest DX9 card to date and if they've done it well, also the fastest card for DX10 when Vista ships. ATI on the other hand can only catch a train for DX10 where it can take the "crown" from NVIDIA. AMD merger stuff isn't really helping them (the ATi), but is certanly helping NVIDIA guys...

I haven't seen this much speculation since... well never

[Razor1]

I haven't seen this much speculation since... well never


Its not really speculation, unless the g80 release date slips up, ATi is going to have to take the crown from nV for dx10 parts, and it will have to do it with a sizable performance advantage.

[bobthebub]

Its not really speculation, unless the g80 release date slips up, ATi is going to have to take the crown from nV for dx10 parts, and it will have to do it with a sizable performance advantage.


For the 'DX10 Performance Crown' to be meaningful there would need to be a benchmark i.e. game, how can dx10 performance be measured at all if a final vista hasn't shipped when these cards start to appear?

[Razor1]

For the 'DX10 Performance Crown' to be meaningful there would need to be a benchmark i.e. game, how can dx10 performance be measured at all if a final vista hasn't shipped when these cards start to appear?


Crysis :grin:

Well ment performance crown in general not just dx10. Developers will be making dx10 games on nV hardware before ATi, the only way ATi will be able to stomp this out is to have a large performance gap, added to this, ATi has to come out before Vista too, otherwise they will loose first sales to nV also. Well they will loose that anyways because of nV's earlier release but I'm sure there will be a surge of discrete graphics card buyers right at and soon after the release of Vista.

[karlotta]

For the 'DX10 Performance Crown' to be meaningful there would need to be a benchmark i.e. game, how can dx10 performance be measured at all if a final vista hasn't shipped when these cards start to appear?One would assume that the Onion will be first out the gate. Though some updated shader/fillrate app might beat it by a nose. And i would go on to say that the big dev/pub houses already know what a ATI dx10 hw can do. To thin k the "g80" is coming out in 2006 is realy pie in the sky. A "g80" thats really a g7x+ will, but a "true" g80/dx10 complete card no.

[trinibwoy]

To thin k the "g80" is coming out in 2006 is realy pie in the sky. A "g80" thats really a g7x+ will, but a "true" g80/dx10 complete card no.

Why not?

[DmitryKo]

OK...

There's NO DirectX 10

:lol:

[Chalnoth]

I don't think semantic arguments as to the name of Microsoft's next 3D API are very useful. Microsoft has changed their naming conventions a few times before. Back in DX9, they decided against calling the graphics package D3D9. Instead they called it DirectX Graphics. Now they're embracing Direct3D and dropping DirectX?

Don't believe it. Both names will stick around. For example, check out the logos on the DirectX developer website:
http://msdn.microsoft.com/directx/

[INKster]

I don't think semantic arguments as to the name of Microsoft's next 3D API are very useful. Microsoft has changed their naming conventions a few times before. Back in DX9, they decided against calling the graphics package D3D9. Instead they called it DirectX Graphics. Now they're embracing Direct3D and dropping DirectX?

Don't believe it. Both names will stick around. For example, check out the logos on the DirectX developer website:
http://msdn.microsoft.com/directx/

Yeah, even during the "WGF 1.0/2.0" era, i've never had any trouble calling them for what they were: DX9 and DX10.
Short, simple.

Recently MS has been trying to apply "Windows" and "Live" on almost any piece of software that they can think of, and naming choices are becoming even more questionable.
What's the problem in following Apple on such mundane aspects ?

[Razor1]

OK...

There's NO DirectX 10

:lol:


LOL too lazy to type out d3d10 :lol:

[overclocked]

One would assume that the Onion will be first out the gate. Though some updated shader/fillrate app might beat it by a nose. And i would go on to say that the big dev/pub houses already know what a ATI dx10 hw can do. To thin k the "g80" is coming out in 2006 is realy pie in the sky. A "g80" thats really a g7x+ will, but a "true" g80/dx10 complete card no.

While im holding my thumbs for ATI to "get in the game more" i think nVidia track-record combinned with its effective and aggresive marketing is hurtening more and more.
It was that damn R520 delay that caused it but i also question some of the choises for time
to have a card.

R580 is forward thinking but G71 draws less power and is alot smaller and in many cases faster.
Now hopefully the forwardthinking buildstones and technology thats in the R5xx series is the ground for R600 and i hope they have succes with that part.

But to count out G80 would be a big misstake imo.

[Chalnoth]

While im holding my thumbs for ATI to "get in the game more" i think nVidia track-record combinned with its effective and aggresive marketing is hurtening more and more.
Well, when I went to nVidia a few months back, I talked to a couple of guys who had previously worked for ATI. Their impression was that nVidia had a company culture that was much more customer-driven, whereas ATI had a culture that was much more technology-driven. That may be a big reason why nVidia seems to be pulling so far ahead with products that really aren't better in many ways.

[INKster]

Well, when I went to nVidia a few months back, I talked to a couple of guys who had previously worked for ATI. Their impression was that nVidia had a company culture that was much more customer-driven, whereas ATI had a culture that was much more technology-driven. That may be a big reason why nVidia seems to be pulling so far ahead with products that really aren't better in many ways.

Companies exist to sell and profit from it, not to create awesome technology that no one buys.
Creating technology is merely means to an end ("To Sell").
Even ATI is like that, no mather what their marketing team says...

Having said that, i can understand that the general perception is that ATI has a more advanced (at least on paper) architecture at the moment, but, just like marketshare, both companies trade blows in that respect too.
Remember R420 (seen as a Radeon 9800 x 2) vs NV40 (first to market with SM 3.0, Purevideo, SLI, etc) ?

Just market cycles, i guess.

[Razor1]

good points also nV just excutes better in everything they do. They have the answer at the right time at the right place where it hurts the most. Like laptops for instance, ATi had the entire pie for add in cards for notebooks, and nV just took a big chuck of that with no response from competition. It taking a cycle possibly even a generation for ATi to get on the ball and start upping the anty which is just too long, once a competitor has a foot hold its easier for them to gain marketshare.

[Geo]

This stuff all tends to be trailing indicators tho, frankly. I don't know too many who weren't hardcore green who were predicting the great NV turnaround in Feb of 2004. Or similarly, non hardcore-red that were predicting the great ATI turnaround in June of 2003. By the time the community is going "woot! woot!" all the hard work is a year old, at least.

[Razor1]

that is true but each company is gauging each other very well, when it comes to performance which is very suprising. Well other then the mishap with the r520. Although the r580 is more technically advanced then the g71 it sure is very unbalanced which is very unusal, pretty much wasted resources, which costs alot of money in the short run and might not be as beneficial as it would have been in the long run since now competition can adapt.

[SugarCoat]

To be precise, there's no such thing as DirectX 10. There is Direct3D10 (which is the name that have always been used at the API level) and anything else in the DirectX runtime is being deprecated. So basically the only official sticker available is "Designed for Windows Vista", though I can see how developers and vendors will still be touting "DirectX 10" or even "DirectX Graphics 10" etc.

Only a few months ago the group manager of windows games and graphics (Chris Donahue) was calling it DirectX 10, if there was ever a sign that name still exists thats good enough for me. I think people are getting too picky over the name, even snooty to keep "correcting" people, there is no issue with calling it DirectX 10.

[neliz]

I think people are getting too picky over the name, even snooty to keep "correcting" people, there is no issue with calling it DirectX 10.

Okay, Microsoft now officially refers to it as "Direct3D 10 system" but that doesn't make it easier, especially since on of the .h files is called d3dx10.h

Also in the SDK, some functions refer to d3dx10, can we assume there is Direct3D 10 and Direct3D X 10?

PIX currently does not capture D3DX10 API calls. The Direct3D 10 API calls that the D3DX10 APIs make will be captured and can be replayed.

[Fdooch]

Okay, Microsoft now officially refers to it as "Direct3D 10 system" but that doesn't make it easier, especially since on of the .h files is called

Also in the SDK, some functions refer to d3dx10, can we assume there is Direct3D 10 and Direct3D X 10?

LOL =)
D3DX is a helper library for D3D. It has functions to work with textures, meshes, vectors, matrices and so on...

[DmitryKo]

I don't think semantic arguments as to the name of Microsoft's next 3D API are very useful. It's about what label is going to be placed on the box.
check out the logos on the DirectX developer website:http://msdn.microsoft.com/directx/ Yes, there's a possibility that MS will call a set of gaming APIs in Vista simply DirectX 10, because that's an established "brand" and will make life simpler for vendors (though "DirectX 10" runtime would probably contain only Direct3D10).

However, some pictures with "DirectX" and "10" are not enough to make such an assumption... it's probably just a shortcut for "DirectX SDK with Direct3D10" ;)

I think people are getting too picky over the name, even snooty to keep "correcting" people, there is no issue with calling it DirectX 10.
See above.

Also in the SDK, some functions refer to d3dx10, can we assume there is Direct3D 10 and Direct3D X 10?The X stands for D3DX (D3D Extensions), an utility library.

[neliz]

See.. that's what I mean, confusing :(

[SugarCoat]

When someone says DirectX 10 you know what they're talking about and 99% of the time it simply doesnt matter if its Direct3D 10 or DirectX 10. For example if someone were talking simply about "those DX10 graphics/features" or "DX10 games" we all know immediatly what they're referring too, no need to incessentaly correct people thats all because the correct term has zero bearing on the topic at hand normally.

That said they may break from the name once vista ships in all forms, formal and informal because DirectX has usually represented backwards compatability where as "DX10" is a whole new animal seperate from all other versions both in software and hardware performance. Or they can continue to be confusing which is also likely as well.

[DmitryKo]

no need to incessentaly correct people thats all because the correct term has zero bearing on the topic at hand normallyI don't know how can I reiterate my point with even simpler wording than what I used above. Sorry.

[chavvdarrr]

http://www.theinquirer.net/default.aspx?article=33784
64 pipes in R600?

[nAo]

Better have a 512 bit bus then..:)

[_xxx_]

full sixty four physical pipes.

:lol:

'nuff said.

[Geo]

Better have a 512 bit bus then..:)

For a quadrupling of ROPs? Shouldn't we be looking for 1024 bus then? :razz:

[Ailuros]

What's there supposed to be so new about R6x0 having (at least) 64 GP-ALUs? :roll:

[Geo]

Nothing, but that's not what he said. He's clearly implying 64 ROPs as well.

[Razor1]

:lol:

'nuff said.


What is Faud smoking lol, I think he think the r600 with have 3 alu's per pipe lol

[caboosemoose]

Nothing, but that's not what he said. He's clearly implying 64 ROPs as well.

What he said may imply 64 ROPs. But if you think he actually understands the implications of what he said, well, you give him much, much mroe credit than I.

[Ailuros]

Nothing, but that's not what he said. He's clearly implying 64 ROPs as well.

Unless you folks are smoking the same stuff as Fudo, I don't see anything that suggests necessarily 64 ROPs. I guess it would be wiser to ask first Fudo what's a pipe exactly to him and how much crack he usually inserts :D

[_xxx_]

with 64 physical pipes, it's sure to blaze through Rydermark!!!

[Subtlesnake]

Companies exist to sell and profit from it, not to create awesome technology that no one buys.
No, but a company may focus its resources on making the right technological call, creating products that represent the best value for the consumer in the long run, or it may focus its resources on making an architecture that represents the best marketing trade-off.

Having said that, i can understand that the general perception is that ATI has a more advanced (at least on paper) architecture at the moment, but, just like marketshare, both companies trade blows in that respect too.
Remember R420 (seen as a Radeon 9800 x 2) vs NV40 (first to market with SM 3.0, Purevideo, SLI, etc) ?
Yes, but that's also a direct reflection of the fact that ATI are a technology focused company. They looked at SM3.0, decided that they wouldn't be able to do it well on 130 nm (i.e have good dynamic branching performance), and that it wasn't significant for the games at the time (arguably the correct call). Instead they implemented DirectX 9.0b which allowed for longer shader lengths, so for example, Crytek could convert their multipass lighting shaders into single pass shaders on ATI's SM2.0 hardware, just like they could with Nvidia's SM3.0 hardware.

Nvidia on the other hand, trumpeted SM3.0 as a major selling point and a huge advantage in games (remember the misleading Far Cry shots?), and were very proficient in persuading gamers that this was the case.

With SLI, I think ATI made a similar mistake. The underestimated the consumer significance of being able to add an extra card into a machine. They didn’t think it was a worthwhile trade-off because performance scales better on single card solutions – two 6800 Ultras are slower than a single 7800 GTX. So they criticised the idea conceptually, but after the success of SLI they had no choice but to follow Nvidia.

Fast forward to the release of ATI's X1800, a chip with vastly better dynamic branching performance - did most gamers know about it, was it widely publicised? No.

Again with the HDR+AA support - It was mentioned in the ATI sell sheets, but only in passing, as a technical feature. Now most gamers do know about this advantage, but I think that's largely due to the reviewers themselves. I haven't seen any publicised ATI material that aggressively trumpets this feature.

[Geo]

Unless you folks are smoking the same stuff as Fudo, I don't see anything that suggests necessarily 64 ROPs. I guess it would be wiser to ask first Fudo what's a pipe exactly to him and how much crack he usually inserts :D

No?

We previously wrote that the chip will have sixty four Shader units but we never realised at the time that the design is actually built around a full sixty four physical pipes.

What goes into a "full. . . physical pipe" that isn't a shader unit that qualifies it to be a "real pipe" (i.e. the article title)? "Real pipe" is clearly a historical referent, isn't it?

[Dave Baumann]

"Pipe" can merely be an implication of parallelism of n items being worked on; n ALU's isn't necessarily so.

[Ailuros]

Fast forward to the release of ATI's X1800, a chip with vastly better dynamic branching performance - did most gamers know about it, was it widely publicised? No.

Again with the HDR+AA support - It was mentioned in the ATI sell sheets, but only in passing, as a technical feature. Now most gamers do know about this advantage, but I think that's largely due to the reviewers themselves. I haven't seen any publicised ATI material that aggressively trumpets this feature.

On both accounts above, it takes a certain amount of time for developers to catch up with specific features/functionalities.

SM3.0 was introduced from NVIDIA in 2004, with no ability to combine float HDR with MSAA and no worth mentioning performance increase through dynamic PS branching; R5x0 appeared in late 2005. Irrelevant of which vendor took which design decisions, wouldn't it be reasonable to assume that if the R5x0 would had apppeared in mid 2004 and the NV40 in late 2005, that the picture today would be completely different also?

[Geo]

Well, it's the "real" and "full" that jangled me. Both are historical referents. The full can't be "full 64", as 64 is already previously in the sentence. "Full. . . pipe" would seem to be the point.

For the record, I am not expecting 64 ROPs. Just engaging in the time-honored B3D parlor game of "roll your eyes at Fudo". :smile:

[Ailuros]

What goes into a "full. . . physical pipe" that isn't a shader unit that qualifies it to be a "real pipe" (i.e. the article title)? "Real pipe" is clearly a historical referent, isn't it?

And what tells me that he isn't on the other hand referring to a SIMD channel or even a TMU? Same "historical referent" applies here to, depending on what anyone called anything in the past.

If you want my own personal speculation I'd rather bet on a 4-way/16x SIMD array than anything else; 16 "pipes" for Zs and 16 "pipes for Ts :P

[_xxx_]

"Pipe" can merely be an implication of parallelism of n items being worked on; n ALU's isn't necessarily so.

Yeah but what about "physical" pipe?

[Geo]

If you want my own personal speculation I'd rather bet on a 4-way/16x SIMD array than anything else; 16 "pipes" for Zs and 16 "pipes for Ts :P

And I'd say that is solidly smack-dab center of the conventional wisdom right now.

[jb]

Yeah but what about "physical" pipe?

Duh, its meants its got 64 pipes to process physics data. I mean with Boundless gaming and all of these other marketing terms about physics being the in thing, then I am sure that what he ment :) :) :) :) :)

[Subtlesnake]

On both accounts above, it takes a certain amount of time for developers to catch up with specific features/functionalities.
But don't you think ATI should be aggressively marketing the advantages it does have, right off the bat? If consumers are aware of and expect developers to take advantage of certain hardware features then surely that actually accelerates the rate of adoption.

I appreciate that adoption was a factor in the appeal of SM3.0, and developers did start to take advantage of it, more than they've taken advantage of the R520's dynamic branching performance, but I wouldn't underestimate Nvidia's role in marketing its advantages in the first place.

SM3.0 was introduced from NVIDIA in 2004, with no ability to combine float HDR with MSAA and no worth mentioning performance increase through dynamic PS branching; R5x0 appeared in late 2005. Irrelevant of which vendor took which design decisions, wouldn't it be reasonable to assume that if the R5x0 would had apppeared in mid 2004 and the NV40 in late 2005, that the picture today would be completely different also?
I agree, but that doesn't invalidate either vendors' design decisions, which were made within the context of what was achievable within that timeframe.

[_xxx_]

Well they were unlucky with all the R520 delay. Then as it came out it was just simply too late to pull it off properly. And since there were pretty much no games out there showing significant advantages thanks to dynamic branching, they had little material to show off.

I think the next round (G8x vs. R6xx) will be much more interesting.

[Razor1]

"Pipe" can merely be an implication of parallelism of n items being worked on; n ALU's isn't necessarily so.


This is what confuses me Dave, how can you have a unfied shader with more then one ALU per shader array working on different shaders at the same time?

What I'm saying is it possible to lets say have the shader array working on one vertex shader instuction and also on a pixel shader instruction?

[trinibwoy]

What I'm saying is it possible to lets say have the shader array working on one vertex shader instuction and also on a pixel shader instruction?

One of Nvidia's recent patents did not discriminate between the thread types that could co-exist in the pipeline of a given ALU. You could have instructions from a pixel thread and instructions from a vertex thread being processed by different stages of the same ALU. The ALU in that case is just viewed as a bunch of available resources and the threads are just a bunch of clients - they just queue up and wait for ALU resources to become available.

[Razor1]

One of Nvidia's recent patents did not discriminate between the thread types that could co-exist in the pipeline of a given ALU. You could have instructions from a pixel thread and instructions from a vertex thread being processed by different stages of the same ALU. The ALU in that case is just viewed as a bunch of available resources and the threads are just a bunch of clients - they just queue up and wait for ALU resources to become available.


Ah that is the answer I was looking for! :lol: Thank you triniy.

Second question :) is the xenos capably of this also? And or does ATi have a similiar technology? Now ATi classifies the 48 shader array xenos as 16x3 quad arrangement. So I still can't see how you can have multiple ALU's of this type per array and still call it 48 just using this as an example. lets say you have 2 of these ALU's per array, wouldn't it come out to something like 32 x 3? Why can't they just come out with a simple way of classification lol!

Also thinking about it somemore, pretty much that patent, is for a unified shader, but lets say you have two of those ALU's in a single array, can they function independently of each other in any order or will there be any possible lock ups like we see with the g70 with the second main ALU?

[Dave Baumann]

This is what confuses me Dave, how can you have a unfied shader with more then one ALU per shader array working on different shaders at the same time?

What I'm saying is it possible to lets say have the shader array working on one vertex shader instuction and also on a pixel shader instruction?

Well, no current design operates with different shaders on different "pipes". Take G7x, for example, we consider that as "24 pipes", because it is operating on 24 pixels in parallel, but the lowest granularity per shader possible is a quad - we don't consider it to be 6 pipes, do we?

[Razor1]

Well, no current design operates with different shaders on different "pipes". Take G7x, for example, we consider that as "24 pipes", because it is operating on 24 pixels in parallel, but the lowest granularity per shader possible is a quad - we don't consider it to be 6 pipes, do we?


ah ok can't forget about history when thinking of the future ;) I hear ya :lol:

[Dave Baumann]

Now ATi classifies the 48 shader array xenos as 16x3 quad arrangement.
Actually, Xenos is clasified as having "3 SIMD Arrays", each one of those arrays is built of 16 ALU's, in a 4x4 quad arrangement - i.e there are 16 pixels (or vertices) operating in parallel within that "SIMD"

[Razor1]

ok hmm so lets say a chip similiar to the xenos is ported over to pc, 3 simd arrays with 16 shader ALU's each, if each shader ALU is doubled, then the classification will be 6 SIMD array's right? I'm trying to line up the rumored 64 shader arrays to this line of thinking, and its not happening lol. Trying to see if ATi can go with a something like a 4 arrays with 32 shader ALU's each array. Is there any limitations or performance advantages of different number of ALU's per array?

what would implications be of have such a large array? Other then major performance cut for "bad" chips.

Added to this the r600 if only a 64 shader alu chip, the g80 might be only a 24 pipeline chip with increased multithreading with 2 alu's each pipe. Trying to think of competitive product's, but rumored specs and performance increase doesn't aline with a 24 pipeline g80.

[ERK]

And what tells me that he isn't on the other hand referring to a SIMD channel or even a TMU? Same "historical referent" applies here to, depending on what anyone called anything in the past.

If you want my own personal speculation I'd rather bet on a 4-way/16x SIMD array than anything else; 16 "pipes" for Zs and 16 "pipes for Ts :P

Ok, let me drag you out a bit further. ;) (or anyone else who wants to respond.)
IYO, will the SIMD arrays contain Xenos-like ALU-pipes, or will they be more like X1X00 pipes with one full, one mini ALU? (Sorry to use the word 'pipe' in this context, but 'ALU' seems similarly ambiguous.) The Xenos-style seem easiest to design (since it is basically still Xenos, just with an extra array), but I was under the impression that the X1X00 pipeline was better performance for PC apps. ?

Also, is there consensus yet that the TEX arrays will also be like Xenos and have 16 bilinear as well as 16 additional point-sampling units? (If so, to what extent can we say it has 32 TEX units?)

Sorry if this has already been gone over...I've been away.:(
ERK

[DmitryKo]

If you look at http://en.wikipedia.org/wiki/Radeon_R520#Chipset_table, R580 is 16:48:16:8 of what hey call "Note 1. (Texture Units:Pixel Shaders:ROPs:Vertex Shaders). All chips of this generation have 1 texture mapping unit (TMU) per pixel pipeline" (and then "Texture" Units gets linked to http://en.wikipedia.org/wiki/Graphics_pipeline :lol:)

Now if we assume Fuad was consulting with this article (I bet he was), we are free to interpret it that R600 configuration is 64:64:32 or 64:192:32 or 64:64:64 or 24:64:24:16 or whatever-combination-of-64-and-random-numbers (in terms of PIPE:ALU(unified):ROP)... which still does not get us anywhere :roll:

[Dio]

Well, no current design operates with different shaders on different "pipes". Take G7x, for example, we consider that as "24 pipes", because it is operating on 24 pixels in parallel, but the lowest granularity per shader possible is a quad - we don't consider it to be 6 pipes, do we?
Well, I made a vague attempt to get people to abandon 'pipes' back in the R300 days, when I was suggesting people use colour/Z/ALU/texture was a better way of giving simple stats than saying "8 pipes, 2 TMU's."

Took you all a year or so to catch up to that one ;) .

[Ailuros]

But don't you think ATI should be aggressively marketing the advantages it does have, right off the bat? If consumers are aware of and expect developers to take advantage of certain hardware features then surely that actually accelerates the rate of adoption.

I appreciate that adoption was a factor in the appeal of SM3.0, and developers did start to take advantage of it, more than they've taken advantage of the R520's dynamic branching performance, but I wouldn't underestimate Nvidia's role in marketing its advantages in the first place.

I don't think ATI didn't point out the differences upon introduction of the R5x0 series of GPUs. The issue being that any of those points get stronger in any user's perception when he can see the differences in real games.

All I'm trying to say is that if R5x0 would had appeared sooner than it did, that we'd have more developer support for the additional features and the end user would see far more differences than today.

It takes years to develop a popular game title. How many new games have you seen recently that actually show clearly said advantages?

[Ailuros]

Ok, let me drag you out a bit further. ;) (or anyone else who wants to respond.)
IYO, will the SIMD arrays contain Xenos-like ALU-pipes, or will they be more like X1X00 pipes with one full, one mini ALU? (Sorry to use the word 'pipe' in this context, but 'ALU' seems similarly ambiguous.) The Xenos-style seem easiest to design (since it is basically still Xenos, just with an extra array), but I was under the impression that the X1X00 pipeline was better performance for PC apps. ?

Also, is there consensus yet that the TEX arrays will also be like Xenos and have 16 bilinear as well as 16 additional point-sampling units? (If so, to what extent can we say it has 32 TEX units?)

Sorry if this has already been gone over...I've been away.:(
ERK

64 ALUs/16 ROPs just comes in handy with the recent rumours of 4 ALUs per pipe. I've said it again that without knowing what each unit is exactly capable of (both for R6x0 as for G80), any of those number speculations are completely senseless.

Assume R6x0 has truly 64 GP-ALUs, then in my mind single issue ALUs sound a bit "weak" for a high end D3D10 GPU, but that doesn't mean that I'm right or wrong either. Or to lay it down in silly speculative math:

64 * 8 FLOPs * 0.7 GHz = 358 GFLOPs
64 * 12 FLOPs * 0.7 GHz = 538 GFLOPs

There could be more FLOPs per ALU, as well as a higher core frequency. Point being that the second case sounds closer to what I'd expect from D3D10 GPUs.

[PeterAce]

ok hmm so lets say a chip similiar to the xenos is ported over to pc, 3 simd arrays with 16 shader ALU's each, if each shader ALU is doubled, then the classification will be 6 SIMD array's right? I'm trying to line up the rumored 64 shader arrays to this line of thinking, and its not happening lol. Trying to see if ATi can go with a something like a 4 arrays with 32 shader ALU's each array. Is there any limitations or performance advantages of different number of ALU's per array?

what would implications be of have such a large array? Other then major performance cut for "bad" chips.

Added to this the r600 if only a 64 shader alu chip, the g80 might be only a 24 pipeline chip with increased multithreading with 2 alu's each pipe. Trying to think of competitive product's, but rumored specs and performance increase doesn't aline with a 24 pipeline g80.

http://www.beyond3d.com/forum/showpost.php?p=767124&postcount=104

[ANova]

Late to the party but my guess would be on 20|64|20 (Texture Units|Pixel Shaders/Vertex Shaders|ROPs) and probably clocks around 800/1200.

[Razor1]

http://www.beyond3d.com/forum/showpost.php?p=767124&postcount=104


Ah thank you thank you ;)

[INKster]

Late to the party but my guess would be on 20|64|20 (Texture Units|Pixel Shaders/Vertex Shaders|ROPs) and probably clocks around 800/1200.

64 shaders, divided between Pixel, Vertex and Geometry ?
I don't know. That seems kind of limited, although i'm not yet aware of how different those units are from the ones already on R580 (minus the Geometry Shaders, of course).
More so if the 600M transistors/80nm rumours indeed hold true.
At least some of the software overhead *should* be removed due to the new Direct3D 10 API, i hope.

[Ailuros]

A lesson I personally learned with the current generation of GPUs (while struggling to understand the theoretical mambo-jumbo for each), is that counting ALUs without knowing what each of them is capable of is completely senseless.

Take the G7x line for instance; I can count 24 dual-issue ALUs with 16 FLOPs throughput per ALU or I can count 48 ALUs with 8 FLOPs throughput per ALU (minus a persentage for texture OPs in both cases).

Nothing tells me anything yet, until I can see what each unit is capable of. I don't see (and that's just a weird example alligned to the one above, which doesn't necessarily make sense) why a 64 ALU config is "underwhelming" if I knew that each ALU is capable of 16 FLOPs, as much as I cannot see why a 128 ALU config would be "overwhelming" on the flip side of the coin if each can do "just" 8 FLOPs.

I don't know what is what yet, neither for R6x0 nor for G80, yet anything below 450-500GFLOPs theoretical throughput sounds low in my mind for a D3D10 GPU. Now anyone of course can do some speculative backwards math trying to figure out the unit amount and each unit's capabilities, yet it is somewhat moot since it would again mean only theoretical maximum throughputs. Granted twice as high as former generation incarnations, but let's not forget the gap between theoretical maximum numbers and real time throughput.

[Geo]

Damn, more than three weeks since a post on this thread. C'mon, ATI, quit skulking in the weeds! WE WANT OUR R600 RUMOURS! :grin:

[SugarCoat]

Good i'm not the only one wondering what the hell is going on when we're suppose to be only a month or two away from launches. Wheres the leaked die shots, news of Crossfire v2 and new PCB info.

[Geo]

Good i'm not the only one wondering what the hell is going on when we're suppose to be only a month or two away from launches. Wheres the leaked die shots, news of Crossfire v2 and new PCB info.


Maybe they want to get RV570 out of the way first. Yeah, yeah, that's the ticket. (I hope).

[SugarCoat]

"what should be do about this R599?"
"We'd like to launch soon, just a few more melting issues to solve."
"We're waiting on the power supplies. *smug experession*"
"The horror.....the horror."
"Fk'in Vista!!!!"


more miffed about the Geforce 8800. Thought they were aiming for back to school sales.

[Geo]

more miffed about the Geforce 8800. Thought they were aiming for back to school sales.

Well, to be fair, while NV made hopeful sounds about September, their bottom line was "before the end of the year". If anyone can point at ATI-on-the-record giving a date range for R600, then I missed it.

[kyetech]

Please dont get my hopes up, by forcing this thread to the top :-) I was so dissapointed not to see any info !! ;-)

[rwolf]

Well, to be fair, while NV made hopeful sounds about September, their bottom line was "before the end of the year". If anyone can point at ATI-on-the-record giving a date range for R600, then I missed it.

I am betting on May next year.

[N00b]

I am betting on May next year.Much too late, IMHO. Vista will be generally available end of january, it seems. So february is my guess with a paper launch a little earlier if neccessary.

[_xxx_]

Hey Geo! Isn't it time for the thread to become "new and improved" too? :)

[LeStoffer]

I am betting on May next year.

Nah, they will try to get it out in December, but it will probably be difficult to find one beyond the initial launch, so the chances of getting one i January might not be much better that jumping on the very first preorder from a decent retailer.

I'm beginning to wonder whether it is at all smart to aim for one of these first DX10 cards these year. The refresh on the 65nm process looks more and more compelling, but I need a new card within the next two months. :sad:

Hmmm, the price on the new 1950XTX looks mighty sweet here in Denmark, so maybe that will be it?

[Geo]

Hey Geo! Isn't it time for the thread to become "new and improved" too? :)


Probably, but I don't have time to do it today. I'm off for a few days here shortly. If one of the other guys doesn't jump on it, remind me over the weekend or next week. :smile:

[Jawed]

http://www.theinquirer.net/default.aspx?article=34373

Jawed

[LeStoffer]

http://www.theinquirer.net/default.aspx?article=34373

Jawed

Yep, just saw that. It doesn't sound good at all. :sad:

theinquirer.net doesn't mention what kind of source they got, so we can hope that it is just a guess on their part.

[Chalnoth]

I thought we've been expecting to not see ATI's R600 until next year for some time now?

[LeStoffer]

I thought we've been expecting to not see ATI's R600 until next year for some time now?

Nope, I was expecting late November or maybe early December. :smile:

[Rangers]

Funny I was arguing with this guy why ATI's R520/80 dies were so big. One of his excuses was they were "building for the future" with all that extra silicon.

So I said, surely their next gen card will come out well before Nvidia's then because they have built for the future, right?

heh.

[Chalnoth]

Well, ATI's next gen coming out before nVidia's should have always been dubious, considering how much younger ATI's R5xx architecture is compared to nVidia's NV4x.

[Rangers]

A guy at Hard claims he heard AMD is going to pull ATI out of the high end GPU market once the acquisition is complete.

Is that true?

Really doesn't look good for ATI.

[Chalnoth]

I doubt that AMD would do something that drastic so suddenly. I think it will happen, personally, but it will be much, much slower.

[rwolf]

After Nvidia is wiped out by integrated cpu/gpu combinations - right Chalnoth? :wink:

[rwolf]

Well, ATI's next gen coming out before nVidia's should have always been dubious, considering how much younger ATI's R5xx architecture is compared to nVidia's NV4x.

Do you think that Nvidia will bring out a rehash of the NV4x architecture to compete against a totally new architecture? I think that ATI might have a leg up because the R5XX architecture included some goodies that will be included with R600. For example I believe, and you can correct me if I am wrong, that the R5XX memory controller supports virtual memory. This means that ATI could have the drivers for this component of Vista all written, tested, and tuned before the R600 comes out.

[Razor1]

After Nvidia is wiped out by integrated cpu/gpu combinations - right Chalnoth? :wink:


Its more about AMD's business focus, I also think it will happen gradually, but it will highly depend on ATi's success with future gpu's if they do well, there will be no change, if they don't there will be some kind of shift to higher priorities for AMD.

[Rangers]

Its more about AMD's business focus, I also think it will happen gradually, but it will highly depend on ATi's success with future gpu's if they do well, there will be no change, if they don't there will be some kind of shift to higher priorities for AMD.


I dont see the point in AMD/ATI dropping high end GPU's at all. No vendor has ever survived providing only mid range and below GPU's, it just isn't possible because if you dont keep up with the high end you're going to fall behind technologically to the point of no return, look at S3 or anybody like that (that is the most profitable segment as well so it doesn't even make business sense).

So that would leave ATI to do what for AMD? Build chipsets? AMD already does that without ATI. There's no need. Maybe IGP's? That's not stuff that's so difficult a expensive acquisition needed to be made for.


I hope Intel steps up if that's the case. One GPU player is going to put us at the mercy of Nvidia.

[_xxx_]

A guy at Hard claims he heard AMD is going to pull ATI out of the high end GPU market once the acquisition is complete.

Is that true?

Really doesn't look good for ATI.

Shifting to another business area is nothing that "doesn't look good". It might not look good to us as gamers for obvious reasons, but if they make profit (be it with high-end gfx or potato chips with new flavour) all is fine for them.

[^M^]

A guy at Hard claims he heard AMD is going to pull ATI out of the high end GPU market once the acquisition is complete.

Is that true?

Really doesn't look good for ATI.

Why would they do that ?
That's fear speaking.

An ADM exec tells that the focus will be on integrate business (they will market it hard) and we are set for months of rumors ...

[vertex_shader]

http://www.ocworkbench.com/ocwbcgi/newspro/viewnews.cgi?newsid1158301575,93080,
"ATI RV600, the next BIG thing

Next month, ATi will launch two cards X1650XT and X1950Pro. Right after that, we should be expecting a new card based on the RV600 core.

From our friends who saw it, the size is like the 7900GX2. The length is 33cm and takes up the space of 3 PCIe slots. It uses a single GPU and require 200W with a 8 pin power connector."

V is sign for value chip, not for highend chip :grin:

3 pcie slot sounds funny too (or this is the next step?), maybe 2slot the card and one is the card own PSU, what take the room for one pcie slot, and not using it real.

[_xxx_]

200W and 3 slots sounds like a pile of crap. If it turns out being true, then they're really nuts.

[Razor1]

200W and 3 slots sounds like a pile of crap. If it turns out being true, then they're really nuts.


LoL and thats the value line chip, they must have a microwave emitter for the high end :lol:

[Jawed]

Hmm, wonder if it could be like the Sapphire Toxic watercooled card.

http://www.extremetech.com/article2/0,1697,1974912,00.asp

Jawed

[INKster]

Hmm, wonder if it could be like the Sapphire Toxic watercooled card.

http://www.extremetech.com/article2/0,1697,1974912,00.asp

Jawed

That is just a two-slot design, and it is a high-end product.

Could they have made a mistake, naming the high-end R600 as the "RV600" ?

[TurnDragoZeroV2G]

RV560 + RV570 -> RV600?

It's not that hard to do if you don't pay attention, I suppose.

Also, why not take a water cooler like the Toxic, but add a second waterblock so that it fits on a dual-GPU, dual-card configuration? Or the equivalent of the Nvidia dual-GPU cards.

Of course, still sounds like a nice steaming pile in the first place...

[ants]

Three slots is not going to happen, for one thing 3 slots wouldn't allow Crossfire on most of the motherboards out there...

[nutball]

This has to be nonsense. Surely?! If not, it'll be one of the biggest ROFLs since NV30 cooling "solution" and Voodoo power brick.

[jb]

A guy at Hard claims he heard AMD is going to pull ATI out of the high end GPU market once the acquisition is complete.

Is that true?

Really doesn't look good for ATI.

That same user at Hard and at Tech Rept, is so far in the green fanboi camp anything he says about the red team is questionable at best. When he posted his "sorces" for this info, it turn out to be some hardware reviewers guess at what he thought might happen...

[Skinner]

http://www.ocworkbench.com/ocwbcgi/newspro/viewnews.cgi?newsid1158301575,93080,
"ATI RV600, the next BIG thing

Next month, ATi will launch two cards X1650XT and X1950Pro. Right after that, we should be expecting a new card based on the RV600 core.

From our friends who saw it, the size is like the 7900GX2. The length is 33cm and takes up the space of 3 PCIe slots. It uses a single GPU and require 200W with a 8 pin power connector."

V is sign for value chip, not for highend chip :grin:

3 pcie slot sounds funny too (or this is the next step?), maybe 2slot the card and one is the card own PSU, what take the room for one pcie slot, and not using it real.

It says R600 now :D
I guess they read the boards ;)

[Pete]

Assuming R600 isn't due for another 4-6 months and that reported sighting is true, I'm not sure we should worry about how it's configured now. Could it be an engineering sample Frankenstein? Maybe some of us with industry contacts could clue us in on what PCBs and with what coolers early samples are fitted?

200W is pretty frightening, though. That sucker better disable a lot of quads when "idle."

[chavvdarrr]

http://www.theinquirer.net/default.aspx?article=34446
250W ?! Someone somewhere is crazy

[Chalnoth]

My money's on Fuad.

[SugarCoat]

Cant see the major harm if its true. What would worry me are the real consumer chips, mid/low end, having issues as well because those are the people that hate to upgrade. And a PSU replacement to an OEM purchaser is major. Most enthusiasts should have a 500W+ PSU in their computer already which wont have an issue with that kind of load at all. I have a 680W thats been neglected :cry:.

Its been so long i cant remember, what was the jump in consumption on the first DX9 boards? I know the 9800 Pro was about 50Watts, how much were DX8 boards previously?


They wouldnt try pushing 1Gig GDDR4 boards yet i wouldnt think....but
Also wonder what type of GDDR4 they'll be using. Samsung is now listing 1400MHz (2800MHz effective) GDDR4 on their product site so theres a possible cause.

I refuse to believe, if i am to take that number as true, that its all from the core and the addition of DX10. That would be grossly overbalanced.

[Rangers]

I wish it was 500 watts for all I care. Just better be about three times as fast as the next guy.

[SugarCoat]

I wish it was 500 watts for all I care. Just better be about three times as fast as the next guy.

instead of a 5-10% spread from the other guys card that launched months before? :lol:

[Rangers]

instead of a 5-10% spread from the other guys card that launched months before? :lol:

Exactly.

Maybe it could come with home liquid nitrogen system.

[Rys]

250W is an absolute nonsense.

[INKster]

Two of these in a Crossfire setup should be interesting...

If this one is true (as well as the other (http://www.theinq.com/default.aspx?article=34319)), then G80 will be much less taxating on the power/heat bill than R600.

[Geo]

I'm more bummed by the Jan/Feb part, tho not terribly surprised since I was more or less sitting on Dec/Jan of late. Ah well.

[Jawed]

So, how long before we start counting R600 as late? :smile:

Jawed

[Geo]

So, how long before we start counting R600 as late? :smile:

Jawed


What, you're applying for "Official 'Late Nazi'"? I don't even remember us posting that position! :razz:

It's bloody hard to say with R600, as they've never put a date on it in public. This is part of why I've wondered if they see it intimately tied to Vista for some reason --there has been an appearance that perhaps they are unwilling to talk about date ranges so long as they are at MS mercy. The current Jan/Feb thing from Inq certainly does nothing to contradict that. . .

[Matas]

If R600 is going to use 8 pin pci-express connector, so it means that all nowadays PSUs will not be able to handle it, because they simply use 6 pin pci-e connectors?

[INKster]

If R600 is going to use 8 pin pci-express connector, so it means that all nowadays PSUs will not be able to handle it, because they simply use 6 pin pci-e connectors?

I'm sure that in such case the return of the Molex-to-PCIe (http://www.surcouf.com/images/Catalogue/07/80/59/g9598007.jpg) adapters will be a given, at least for a while.
The only difference now being the 8pin connectors, instead of 6.

[Bouncing Zabaglione Bros.]

I'm sure that in such case the return of the Molex-to-PCIe (http://www.surcouf.com/images/Catalogue/07/80/59/g9598007.jpg) adapters will be a given, at least for a while.
The only difference now being the 8pin connectors, instead of 6.

Some people would actually need 6pin->8pin connectors, given the number of PSUs that now offer dedicated rails and plugs just for the PCIe connectors and power draw.

[Matas]

Thanks for the replies. I was getting afraid, that mine Fortron Epsilon 700W won't have a chance to try handle R600...

[LeStoffer]

250W is an absolute nonsense.

Thanks for the "correction". Now I can continue to peacefully dream about The Unspolied Monster! :cool:

[Geo]

Btw, just for funsies. . . any opinions on how far into 2007 R600 would have to slip before 65nm becomes a reasonable possibility? :twisted: Does February make anybody twitch? Or still a little soon?

[trinibwoy]

Wouldn't the process have to mature before they even consider it for such a complex chip? Even if TSMC masters 65nm by then, if ATi have been gearing up for 80nm all this time then it will take a considerable amount of time to go through the whole process again for the 65nm node. So my bet is that R600 has to slip to fall 2007 for us to see it on 65 nm :)

[_xxx_]

So my bet is that R600 has to slip to fall 2007 for us to see it on 65 nm :)

Nonono! Fall 2007 is the time to speculate about R700 and G90 already ;)

[Geo]

Wouldn't the process have to mature before they even consider it for such a complex chip? Even if TSMC masters 65nm by then, if ATi have been gearing up for 80nm all this time then it will take a considerable amount of time to go through the whole process again for the 65nm node. So my bet is that R600 has to slip to fall 2007 for us to see it on 65 nm :)

I dunno, there have been rumors they might have done both --surely they would have been planning a 65nm refresh well before now, worst case.

They didn't go "low/mid first to new process" with R520. . . but then that rather bit 'em in the heinie, didn't it?

[trinibwoy]

I dunno, there have been rumors they might have done both --surely they would have been planning a 65nm refresh well before now, worst case

Yeah I didn't consider that they might be looking ahead to R600 on 65nm already even if the first spell is on 80nm. Even so, is it realistic to expect that progress on that front is so far advanced that it would be no big deal to just hop from 80nm to 65nm given a couple more months? They would need to have 65nm samples in hand pretty soon if they wanted to have volume by February. I would be more optimistic if the current target dates were a result of a 65nm push but we don't have any info to suggest that to date. We also don't have any evidence that TSMC is even capable of putting out R600 on that node in that timeline.

[Geo]

Truthfully, February feels early to me too for 65nm on a chip that complex. I was just curious what others thot on that front, and what month has to pop up as a target before the spidey sense starts trilling its tune.

[LeStoffer]

Truthfully, February feels early to me too for 65nm on a chip that complex.

I highly doubt that the TSMC 65nm process is even mature enough for less complex chips like a midrange DX 10 card. We should count ourselves lucky if it is doable at spring time. I would put my money on second half of 2007.

[Matas]

One more question regarding R600 and his 250W power consumption:
I have FSP Epsilon 700W PSU (4x 12V rails at 18A). So it means that one pci-express rail can handle ~216W (12 x 18A). If R600 uses 250W does it mean that my PSU will not handle it, because at pci-e rail Epsilon can handle only 216W? I'm thinking right or not?

[Jawed]

The PCI-Express slot provides 75W of power in addition to the auxiliary power socket on the card.

Jawed

[Matas]

I'm talking about my PSU's pci-express rail, not motherboard.

[Bouncing Zabaglione Bros.]

Now the Enermax Goliath (http://www.coolergiant.co.uk/products/psu/galaxy/EGA1000EWL/index.asp) makes sense. Watch the lights dim as you turn on that </Dr. Evil> One Thousand Watts </Dr. Evil> of PSU goodness!

[Kaotik]

It's funny how peopel are going on about the watts, like they would really matter.
After all, in modern normal computers, without gazillion HDD's, only thing that matters is the 12V line. For example Be Quiet!, one of _the_best_ brands out there (only available within EU, though, I think), offers 430W and 530W model of certain PSU-line, and really, they're just as good for the usual (gaming) PC - they have 100% identic 12V line(s)

[Jawed]

I'm talking about my PSU's pci-express rail, not motherboard.
They should all be independent of each other and hopefully at least one of them is independent of the motherboard :!: ...

Jawed

[turtle]

They should all be independent of each other and hopefully at least one of them is independent of the motherboard :!: ...

Jawed

Right. A card would require 15a on a 12v rail from a connector (250-75=175/12=15a) if I understand correctly, which seems to be the minimum (on minimum maximum a single line if others are lower) on PSUs with multiple 12v rails. Makes sense.

[Matas]

So, nobody had answered my question...

[SugarCoat]

So, how long before we start counting R600 as late? :smile:

Jawed



I'd go off the Vista launch. No X2000 by Vista launch, i'd say its late. And theres a good chance that may become a reality.


Technically speaking someone could say its already late since they're going to overshoot the back to school season as well as the holiday season, and no company that sells products wants to do that.


One more question regarding R600 and his 250W power consumption:
I have FSP Epsilon 700W PSU (4x 12V rails at 18A). So it means that one pci-express rail can handle ~216W (12 x 18A). If R600 uses 250W does it mean that my PSU will not handle it, because at pci-e rail Epsilon can handle only 216W? I'm thinking right or not?

+75W through the socket which is the primary source every card uses then draws off the 12V for what ever else it needs. The problem in that situation is going to come for PSUs that operate everything off a single 12V line. Some even split a single 12V. Remember kids, buy server rated PSUs since they usually come with at least 4 independent 12V lines. ;).

[turtle]

So, nobody had answered my question...

If you have a rail that is independent of the motherboard connectors (as Jawed said) which you obviously do as probably one is more-than-likely specifically devoted to the ATX connectors, and one of those free lines has at least 15a on it (if the max load is 250w) you'd be fine. If you have 18a on a free rail independent of the motherboard power, you could run up to 291w, 216 from the rail and 75w from the other line that is powering the motherboard.

Short: Answer is yes.

[Matas]

On FSP forums I have found this:
"Using a gfx card without a direct PCI-E connectoer will draw power from 12v3 since it's designed with the slot on MB. The "official" max load on 12V3 is 15A but I was told it can be up to 18A by our head office today. They registered it as 15A when they applied safety certificates but will update it for the next version".

[Chalnoth]

One more question regarding R600 and his 250W power consumption:
I have FSP Epsilon 700W PSU (4x 12V rails at 18A). So it means that one pci-express rail can handle ~216W (12 x 18A). If R600 uses 250W does it mean that my PSU will not handle it, because at pci-e rail Epsilon can handle only 216W? I'm thinking right or not?
Right, so I think this is a strong reason to believe that 250 watts is insane.

[SugarCoat]

Right, so I think this is a strong reason to believe that 250 watts is insane.

On the other hand both ATI and Nvidia have been suggesting a minimum of a 18A 12V line for the last year if not more for their high end cards. Perhaps in preperation? And as already mentioned a couple times now, 18A dedicated line would be more then enough to power 250W, even a 16A dedicated line could get by.

Someones gotta play devils advocate to keep it interesting!


If we ever saw a single card sku with 1024mb of 2800MHz GDDR4, that may also cause a decent spike.

[turtle]

On the other hand both ATI and Nvidia have been suggesting a minimum of a 18A 12V line for the last year if not more for their high end cards. Perhaps in preperation? And as already mentioned a couple times now, 18A dedicated line would be more then enough to power 250W, even a 16A dedicated line could get by.

Precisely. 18a would power up to ~300w, which is what Anandtech rumored the max for the high-end of this next gen would use. I think the 18A suggestion was def in preparation for G80 (perhaps GX2 if G80/G81 use less than 150w) and R600, which could use up to 250w (15A) its self.

@ Matas: I don't know how your rails are set up, but assuming both 6-pin plugs on your psu are on different rails, which they probably are, you'd just need to buy a 6-pin-->8-pin adapter (for one or two cards.)

If both are on the same line, which they probably aren't, if you wanted to run more than one card you'd be forced to use the 12v3 rail for one using a molex --> 8-pin adapter.

This is assuming the 8-pin rumor is true.

[Matas]

But if my pci-e rail isn'i able to take 75W from pci-express slot, it can only handle 216W(12x18A) and in taht way it is not enough for R600.

[SugarCoat]

But if my pci-e rail isn'i able to take 75W from pci-express slot, it can only handle 216W(12x18A) and in taht way it is not enough for R600.

every board is made to deliver 75W through the slot, i dont really see what you're questioning. If your PSU is using the third 12V line for the slot then you simply use one of the other lines to keep a safe margine of error. Unless you're running multiple RAID arrays and some overclocked pentium D beast AND have modified a 12V line to power a TEC for that processor, there shouldnt be any power issues.

[Matas]

Read my written post from FSP forum...

[SugarCoat]

what the 15A official 18A unofficial thing? Why does that matter? I think you're overcomplicating way too much. Let me simply say, your PSU is fine. If you want to doubt that then thats your business.

[turtle]

every board is made to deliver 75W through the slot, i dont really see what you're questioning. If your PSU is using the third 12V line for the slot then you simply use one of the other lines to keep a safe margine of error. Unless you're running multiple RAID arrays and some overclocked pentium D beast AND have modified a 12V line to power a TEC for that processor, there shouldnt be any power issues.

This is true...or reiterated:

You probably have a 12v rail (18A) devoted just to the motherboard through the ATX (20+4pin) and 8-pin plugs. Most of the 12v power from that line is going to go to powering the pci-e slots. If you're using a board that uses 20+4 and then molex connectors, you could use the molexes off different line if you're worried...Which you shouldn't be unless you're running a chipset/cpu that uses a crap load of power. You should have more than enough enough to power on the devoted line to power a couple pci-e slots at 75w. I'm sure you could find a combo between 4x18A that will will work fine to power the board + the cards power connector. I believe this is what Jawed/SugarCoat and others were trying to get across to you.

[turtle]

Apologies on the DP. I would edit my earlier post...but alas no edit function. :razz:

http://theinquirer.net/default.aspx?article=34454

The Inq debunking rumours...That's like a paradox wrapped in a enigma wrapped in a mystery of wtf.

[Matas]

what the 15A official 18A unofficial thing? Why does that matter? I think you're overcomplicating way too much. Let me simply say, your PSU is fine. If you want to doubt that then thats your business.

I don't want to argue.. I just a bit don't understand your explanations. What's all...

[SugarCoat]

I don't want to argue.. I just a bit don't understand your explanations. What's all...

I'll try once more just cause i dont like cluttering this thread.


Your PSU = 4 x 12V lines each rated at 18A. They are independent which means each one can offer you just over 200W. One of them is already powering the motherboard, so for the sake of argument lets simple say that 12V line 3 is not an option to you, it doesnt exist to use for addons or peripherals. Now that 12V line 3 is feeding the motherboard and the PCI Express power through the slot, so that is 75 Watts right there for your card to use thanks to that one dedicated line. The other 3 lines, 12V1,2, and 4, you have can STILL put out just over 200W each, so you could use your 12V4 for your imaginary 250W power sucking/wasting card and that would essentially be like feeding about 280 Watts through 2 seperate lines which your PSU is more then capable of.



The only way you would have an issue, is if your PSU had 1 12V line rated at 18A. Then yes there would be a problem because that 12V is feeding the motherboard and then plugging that into that 250W card would cause an issue since you have no addition of power, 75W through the motherboard and ~200 through the 12V molex are coming from the same line. But your PSU doesnt have that problem, it has 4 seperate lines each capable of delivering their own power. If one is in use for the max output, it doesnt make the other 3 rails useless.

[turtle]

^ Dude, I love ya (in a non-sexual bipartisan friendly sorta-way) but I think we're fighting a lost cause. :lol:

[psurge]

+75W through the socket which is the primary source every card uses then draws off the 12V for what ever else it needs. The problem in that situation is going to come for PSUs that operate everything off a single 12V line. Some even split a single 12V. Remember kids, buy server rated PSUs since they usually come with at least 4 independent 12V lines. ;).

Actually, from what I understand, that's not true at all. Most power supplies have just 1 12V line internally that can provide much more than 18A. Some split them into multiple "independent" rails (so that they can call themselves ATX compliant). But (for example) a PC Power & Cooling Silencer 750 (one 12V rail @ 60amps) shouldn't have much difficulty with a 250W graphics card and dual CPUs.

The only power supplies I know of that have truly independent 12V rails are the PC Power & Cooling TurbCool 850W and 1kW models. Those have 3 completely separate sets of AC to DC conversion circuits - 2 for 12V and 1 for 3V/5V line. They also cost $450-$500.

[SugarCoat]

true but i was talking about worst case scenerio. $50-$80 PSUs usually have a single 18-20A rail, sometimes split so that 18 or 20A is spread across 4 molex. Those seem the most common for computer builders and even OEMs within the last year or two.

[psurge]

I had no idea, that's crazy low. (Dunno why I'm surprised though).

[nutball]

The Inq debunking rumours...That's like a paradox wrapped in a enigma wrapped in a mystery of wtf.

This is hardly debunking the rumours, more like throwing petrol on them...

Graphics are going to get three-slot, and even four-slot cooling, but the cards will not be using PCI Express slots, but rather sockets on the motherboard.

[Neeyik]

Apologies on the DP. I would edit my earlier post...but alas no edit function. :razz:

http://theinquirer.net/default.aspx?article=34454

The Inq debunking rumours...That's like a paradox wrapped in a enigma wrapped in a mystery of wtf.
Kewl. So according to The Inq, the R600 has managed to drop by 70W in less than one day! OMGFATIr0x!!!!!1!!

[Razor1]

Kewl. So according to The Inq, the R600 has managed to drop by 70W in less than one day! OMGFATIr0x!!!!!1!!

LOL, that sounds much more resonable now

[wishiknew]

Maybe The Inq ment 180w for the gpu chip alone and 250w for the whole card.

God I've gone so low as to decipher from that site.

[Razor1]

still can't see 250 watts on one card, man thats alot of juice, my whole system doesn't take more then 200 watts with 7 harddrives, a 4800+ creative sound blaster xfi fatality a 7800 512, 2 dvd burners, (everything going at max and it might just go above 200 at most 210 watts)

[Jawed]

On the other hand both ATI and Nvidia have been suggesting a minimum of a 18A 12V line for the last year if not more for their high end cards. Perhaps in preperation?
18x12=216W - yet X1900XTX under load along with the CPU under load and the efficiency losses of the PSU all amounts to 290W at the wall:

http://www.techreport.com/reviews/2006q3/radeon-x1950xtx/index.x?pg=9

The PCI Express standard says 75W through the mobo + 75W through the dedicated power line.

The CrossFire X1950XTX configuration draws 110W more power than the single card config. Say 120W for a single card running at full clocks (CF down-clocks a bit).

Jawed

[hoom]

Crossfire downclocked an x1900xtx to x1900xt speed because the x1900 crossfire ran at x1900xt clockspeed but x1950 crossfire is same clock speed as the x1950xtx so it shouldn't do any downclocking :???: (ooh say that 10 times)

[_xxx_]

18A dedicated line would be more then enough to power 250W, even a 16A dedicated line could get by.

18A * 12V = 216W :)

EDIT: oops, Jawed beat me to it...

[SugarCoat]

i assumed someone would already add in what the card would draw through the socket because thats a given..

75+216 +/- 5% = .. :wink:

[rwolf]

I highly doubt that the TSMC 65nm process is even mature enough for less complex chips like a midrange DX 10 card. We should count ourselves lucky if it is doable at spring time. I would put my money on second half of 2007.


¡°You don¡¯t want to develop a technology first and then find customers for it,¡± said Kin. ¡°Customers drive
the technology, not the other way around.¡±

That may be why the first 65nm production at TSMC will be for low power devices for the cell phone industry,
a high volume industry with very short product cycles. Those first devices will use a standard low-k gate
dielectric, said Kin. TSMC plans to continue to use Applied Material¡¯s Black Diamond low-k dielectric
material in its copper interconnect process at the new node.

But the second round of 65nm technology, focused on high-speed devices and due in mid-Q3 of 2006
may add a high-k transistor gate dielectric to the mix. ¡°Some of the high performance customers have
expressed interest in that, but there is nothing firm yet,¡± said Kin.

http://www.kasausa.com/?doc=bbs/gnuboard.php&bo_table=Tech&page=1&wr_id=34

[mashie]

250-300W of heat from each videocard will be very interesting to evacuate from the case in a silent fasion.

[RejZoR]

Water cooling serial? :lol:

[SugarCoat]

Actually i was thinking about messing around with a TEC on my next card regardless of power/heat (rumors true or not). Problably save me some space too with the way stock coolers have been growing. I just hate messing around with acrylic lacquer.

[tEd]

http://img71.imageshack.us/img71/4963/4918qo51x81tal4.jpghttp://img71.imageshack.us/img71/2056/u3t9pvpwxd10fa9.jpghttp://img71.imageshack.us/img71/4740/zplaa92921teqi4.jpg

Someone posted these on another forum . Looks not real to me but hey it's a rumor thread.

[Arty]

I thought we were back to R6x0 or has that changed again?

[no-X]

"R600 will double the performance or R580" :roll:

[ants]

Someone posted these on another forum . Looks not real to me but hey it's a rumor thread.

Crappy, they definately look fake.

"R600 will double the performance or R580"

...

EDIT: Beaten! was searching for more spelling/grammatical errors :p

[Pete]

The second slide alone has three spelling/grammatical errors. You'd think a company about to unleash "absolute performance leadership" could afford an editor. :razz:

[INKster]

Yep, looks fake to me too.
The language used is not exactly the kind of polished, somewhat restrained wording common in real marketing materials.

[Arun]

Look at the small symbol on the G80 slide: http://img132.imageshack.us/my.php?image=hardspellslidenj6.jpg
Clearly, it's the same morons spreading the same FUD.

Uttar

[tEd]

at least we are over the watt discussion because it started to annoy me :wink: