NVIDIA GT200 Rumours & Speculation Thread

[Megadrive1988]

I still would not consider Nvidia next GPU to be a true next-gen GPU architecture. Even though it's been so long since the G80 (NV50). The next GPU is still to be considered NV55, not a new clean sheet design (NV60).

While it's expected to bring significant changes and upgrades(ala G70 over NV40), and while it will be called a next-gen GPU by Nvidia, it'll still be using much of the architecture layed down with the G80.

I don't expect NV60, Nvidia's new clean-sheet architecture until late 2009.


so
G80 / GF8800 = NV50
G92 / GF8800 = NV50
GeForce GX2 9800 = NV50


so called "GT200" = NV55 ?

true next-gen = NV60

 

[turtle]

Don't you think that if it was going to be released in may more info would have leaked out by now? We still don't know the slightest thing about how this chip compares to G80.

Except Computex is the during the first week of June, so to me Arun is implying info concerning an early Q3 release, or there-about.

For as long as I can remember we usually get *real* information leaked somewhere around ~2 months, give or take a couple weeks, before launch, or probably when initial production silicon/architecture info/clock estimates is received by someone (usually an AIB I assume) whom chooses to spill the beans. The exception of course being G80, which was a little over a month before for the solid info, but really two months before launch there was that hardspell report that was more-or-less correct, but reported incorrectly because of the major change in architecture (shader clocks, non-uniform memory bus, etc). Anyway, I'd say about two months is normal, slightly more for ATi, slightly less for Nvidia.

So no, I wouldn't be surprised if it was launched in May and we don't know anything yet, but would expect something to leak soon, especially if what we've heard so far is incorrect (384, 512-bit, etc). If it will be shown in June, and launched around that time or perhaps a little later, I wouldn't be surprised if we don't have more concrete info for another month or so.

 

[Domell]

I still would not consider Nvidia next GPU to be a true next-gen GPU architecture. Even though it's been so long since the G80 (NV50). The next GPU is still to be considered NV55, not a new clean sheet design (NV60).

While it's expected to bring significant changes and upgrades(ala G70 over NV40), and while it will be called a next-gen GPU by Nvidia, it'll still be using much of the architecture layed down with the G80.

I don't expect NV60, Nvidia's new clean-sheet architecture until late 2009.


so
G80 / GF8800 = NV50
G92 / GF8800 = NV50
GeForce GX2 9800 = NV50


so called "GT200" = NV55 ?

true next-gen = NV60

Well, i`ve thought NVIDIA dropped "NVxx" name scheme 2 years ago (after releasing G70 aka NV47).

Another thing is if latest rumours about GT200 are true there is no good for NVIDIA. Only 2xFPU performance over G92 will give only some 40-50% performance boost to G92. When we compare it to Rv770 (160SPs!!!! it`s 2,5xRv670 SPs) rumours it seems GT200 won`t have any chance even against next-gen mainstream GPU from ATI. I hope GT200 will be much stronger than rumoured at present

PS. But the most important question is - is GT200 a new name of G100 or both of them are totally different GPUs? I`m asking because on the first page of this topic CJ have said NVIDIAs next-gen GPU will have 512-bit memoty bus. So when he was saying it was he thinking abiut GT200 or G100 (if they aren`t the same GPUs).

 

[XMAN26]

Another thing is if latest rumours about GT200 are true there is no good for NVIDIA. Only 2xFPU performance over G92 will give only some 40-50% performance boost to G92. When we compare it to Rv770 (160SPs!!!! it`s 2,5xRv670 SPs) rumours it seems GT200 won`t have any chance even against next-gen mainstream GPU from ATI. I hope GT200 will be much stronger than rumoured at present

Some how I doubt your assurtions. Right now, the X2 can just match an already 15m old video card. The X2 has 2x 64SP chips compared to a single 128SP G80 core. Your assuming that ATIs 160SP(thats if the 96SP isn't true) will somehow be too much for 256SPs from GT200. I'd agree if, and only if, ATI can some how make those SPs perform better than they do now.

 

[DemoCoder]

I think expecting NVidia to jettison the G80 for a 'clean sheet' is expecting too much. They've already got a very efficient chip, there's probably not much more they can do to boost efficiency, so at this point, I think they're going to simply scale the architecture, add features (DP, 10.1), and optimize costs. It took NVidia 4+ years to bring the G80 architecture to fruitition, so I don't see any clean sheet designs happening so quickly.

NVidia was able to eek alot out of the GF1 architecture, and NV3x, and given the lead the G80 already has, I think they'll play it safe and go with evolutionary changes.

 

[Pete]

Another thing is if latest rumours about GT200 are true there is no good for NVIDIA. Only 2xFPU performance over G92 will give only some 40-50% performance boost to G92. When we compare it to Rv770 (160SPs!!!! it`s 2,5xRv670 SPs) rumours it seems GT200 won`t have any chance even against next-gen mainstream GPU from ATI.

Well, rumors are rumors. Let's not get crazy comparing one rumor to the next, especially considering how often they've changed. But I'm curious why 2x G80 SPs aren't in the same league as 2.5x R600 SPs considering their current relative performance, and why you're focusing on SPs given, well, their current relative performance. Not to mention that it's laughable on its face that NV's next incremental (not a generational leap like NV30) design couldn't compete against ATI's "mainstream" part. I mean, let's not lose all perspective here.

PS. But the most important question is - is GT200 a new name of G100 or both of them are totally different GPUs?

I'm still wondering whether D8E/GT200/G100 are all the same thing, namely NV's next high-end part.

 

[compres]

I believe GT200 was first mentioned by FUDO, so I call bullshit. G100 is the most likely code name IMO.

 

[Slyne]

I'm still wondering whether D8E/GT200/G100 are all the same thing, namely NV's next high-end part.

I was just thinking... maybe Nvidia threw a few names to a few different ppl (AIB?), to get any idea of where its leaks came from?

 

[INKster]

I believe GT200 was first mentioned by FUDO, so I call bullshit. G100 is the most likely code name IMO.

G00613 ?

 

[Domell]

Some how I doubt your assurtions. Right now, the X2 can just match an already 15m old video card. The X2 has 2x 64SP chips compared to a single 128SP G80 core. Your assuming that ATIs 160SP(thats if the 96SP isn't true) will somehow be too much for 256SPs from GT200. I'd agree if, and only if, ATI can some how make those SPs perform better than they do now.

First how do you know how many SPs GT200 has? Second remember that Rv770 IS NOT highend part but GT200 is (at least is supposed to be). So if ATIs mainstream next gen GPU will be as fast as NVIDIA highend GPU then Rv770 is real winner.
There will probably be something like that: GT200 vs r700 and Rv770 vs NVIDIAs mainsteram GPU based on GT200 architecture. So GT200 should be about 40-50% faster than Rv770 at least if NVIDIA doesn`t want to loose their position. So GT200 FPU performance has to be much greater than only 2x.
I hope GT200 significantly faster than Rv770 and it will be able to fight against r700 (rumoured 2xRv770)

 

[pjbliverpool]

First how do you know how many SPs GT200 has? Second remember that Rv770 IS NOT highend part but GT200 is (at least is supposed to be). So if ATIs mainstream next gen GPU will be as fast as NVIDIA highend GPU then Rv770 is real winner.
There will probably be something like that: GT200 vs r700 and Rv770 vs NVIDIAs mainsteram GPU based on GT200 architecture. So GT200 should be about 40-50% faster than Rv770 at least if NVIDIA doesn`t want to loose their position.
I hope GT200 significantly faster than Rv770 and it will be able to fight against r700 (rumoured 2xRv770)

If RV770 is AMD's fastest single GPU solution then its their fastest GPU as far as i'm concerned. Multiple GPU configs don't IMO due to being unreliable and horribly inefficient.

Of course if AMD can build R700 with dual GPU's that don't require a doubling up of memory and offer a 50% of more performance increase across the board without ridiculous power/heat requirements, I may change my mind.

Even then though there is the issue of control lag.

Nope, its single GPU's or nothing for me. I just hope that whatever single GPU AMD comes up with can hold its own against NV's next gen.

 

[XMAN26]

First how do you know how many SPs GT200 has? Second remember that Rv770 IS NOT highend part but GT200 is (at least is supposed to be). So if ATIs mainstream next gen GPU will be as fast as NVIDIA highend GPU then Rv770 is real winner.
There will probably be something like that: GT200 vs r700 and Rv770 vs NVIDIAs mainsteram GPU based on GT200 architecture. So GT200 should be about 40-50% faster than Rv770 at least if NVIDIA doesn`t want to loose their position. So GT200 FPU performance has to be much greater than only 2x.
I hope GT200 significantly faster than Rv770 and it will be able to fight against r700 (rumoured 2xRv770)

Again, I doubt all your assumptions you have as to how the RV770 will perform unless ATI has somehow fixed how poorly they perform now. As it stands now, 64SP(320US) can't compare to 128SP by Nvidia. 2x(X2) ATI 64SP(640US) can just match that same 15m old 128SP by Nvidia. And you are some how coming to the conclusion that 96SP(480US and most likely the correct rumor) or your hopes 160SP(800US) is some how going to destroy the rumored so specs of either 256SP or 384SPs. either of which would work at worst on par to your high hopes situation.

 

[LordEC911]

Again, I doubt all your assumptions you have as to how the RV770 will perform unless ATI has somehow fixed how poorly they perform now. As it stands now, 64SP(320US) can't compare to 128SP by Nvidia. 2x(X2) ATI 64SP(640US) can just match that same 15m old 128SP by Nvidia. And you are some how coming to the conclusion that 96SP(480US and most likely the correct rumor) or your hopes 160SP(800US) is some how going to destroy the rumored so specs of either 256SP or 384SPs. either of which would work at worst on par to your high hopes situation.

Why are you comparing shaders/SPs across different architectures...

 

[XMAN26]

Why are you comparing shaders/SPs across different architectures...

I'm not, Domell is and i'm arguing his thought dont jive with what we know.

 

[compres]

Yeah but your comparison is an oversimplification, IMO ofc.

 

[Ailuros]

First how do you know how many SPs GT200 has? Second remember that Rv770 IS NOT highend part but GT200 is (at least is supposed to be). So if ATIs mainstream next gen GPU will be as fast as NVIDIA highend GPU then Rv770 is real winner.
There will probably be something like that: GT200 vs r700 and Rv770 vs NVIDIAs mainsteram GPU based on GT200 architecture. So GT200 should be about 40-50% faster than Rv770 at least if NVIDIA doesn`t want to loose their position. So GT200 FPU performance has to be much greater than only 2x.
I hope GT200 significantly faster than Rv770 and it will be able to fight against r700 (rumoured 2xRv770)

To answer your question with another question, how on God's green earth do you know that GT200/100 (or whatever it's called) "FPU" performance is 2x times higher?

Assume NV targets X increase of overall performance over today's highest end GPU. One way would be to increase the amount of SPs by Y times and another way would be to increase efficiency per SP by Z persentage. The latter scenario worked fine for comparisons between G71/GX2 and G80. In sterile theoretical maximum numbers the MADD GFLOP performance has exactly how much difference between those two?

All the above shouldn't mean anything; it's merely to show that with as many unknowns all these assumptions are meaningless. I'd personally have a long hard laugh if all so far "GT200" rumours have absolutely nothing to do with what we've read so far.

 

[Squilliam]

How do you guys think this will fit into the Nvidia pricing schema?

My take on this is that if both the 55nm g92 and the gt200 rumours come to fruition. The high end will be the Gt200/100/whatever and the mid/mid-low end parts could be cheap g92 derivatives. Then next generation the gt200 becomes shrunk scaled to mainstream? A similar strategy to what was done with the g80 generation? g92/4=600 series/400 series whilst the new gt200=800 series, whatever monikor they put before it doesn't matter.

 

[XMAN26]

Yeah but your comparison is an oversimplification, IMO ofc.

Again it isn't me, its Domell. He is the one saying that and also expecting some huge massive performance increases from ATI in the RV770 while totally disregaurding how the current SPs from ATIs cards currently compare performance wise to even the G80. All I'm doing is pointing that discrempency out to him. Just read his posts and you'll see what I'm saying.

 

[CJ]

My take on this is that if both the 55nm g92 and the gt200 rumours come to fruition. The high end will be the Gt200/100/whatever and the mid/mid-low end parts could be cheap g92 derivatives.

Bingo.

Jen-Hsung even said at the last conference: "we'll be driving the costs of G92 down over the next couple of quarters" which means that G92 will be around for at least 6 more months and a shrink to 55nm will certainly help driving down the costs of G92 down. GT200 will be the replacement of GF9800GX2 in the highend. G9x stack will fill everything below it... And supposedly they even have something else on top of GT200 too....

 

[hoho]

This sounds quite similar to G80 at its early days when only 8800GTX/GTS were availiable.