NVIDIA GF100 & Friends speculation

[Tahir2]

By the way, why was there no mention of video processing capabilities? Unchanged from GT21x presumably?

Do you think video decoding improvements are interesting anymore as a USP? On the PC space do we need to decode more than 2 1080p streams in hardware or have we past the point of diminishing returns?

 

[FrameBuffer]

Prices and performance has nothing in common. 5670 cost $100 and 5970 cost $700 -- is it 7x faster? No. So does that mean that everyone should go and buy 5670? Nope. Price is what you're ready to pay for a product and wrt graphics cards performance in today's games is not the only factor of pricing. So if 5870 will have 75-80% perfomance of GF380 and will cost 60% of GF380 then that's because GF380 has some other benefits to a buyer beyond performance alone. I've already described some of these benefits. Surely if you don't thik they're important then you're better of buying 5870 -- IF you're OK with it's performance because deltas aren't absolute numbers. If enough people will think the same NV will be forced to drop the prices. So that'll be solved one way or another, so I don't see any reason to talk much about it.

Maybe not 7x, and as we all know there is premium for top "Halo" performance parts. Part of the problem is there is no competition, meaning there is no reason for ATI to drop prices being that they are most likely selling all 5900 series parts at the price the market will bare.

http://www.computerbase.de/artikel/hardware/grafikkarten/2010/test_ati_radeon_hd_5670/5/
shows 5970 is only 5X faster then the 5670. Though I don't doubt under extreme resolutions (2560x1600 8X AA 16AF) the 5970 would perform close to or over 7X faster then the 7X cheaper 5670.
http://www.computerbase.de/artikel/hardware/grafikkarten/2010/test_ati_radeon_hd_5670/13/
for example shows that in Crysis (1680x1050 8xAA/16xAF) is 6.77X faster and most definitely would show a greater perfromance increase at 1900/1920/2560 resolutions.. so to answer your question. Yes the 5970 can be up to 7X faster though it cost 7X as much.

((If anyone can point me a link or 3 showing someone brave enough to run the 5670 @ 1900 - 2560 resolutions with 8X AA 16AF I'd be grateful, and again I don't doubt that the 5970 would show 7X performance increase))

I don't know how it'll end up in the end with the whole Fermi line-up but with GF100 i'm 90% sure that I have a pretty good understanding of what (and when) to expect from a final products.
That's why I'm saying that it's strange to see anyone dissapointed with GF100 graphics architecture info. It looks like people are dissapointed with not knowing fps numbers but somehow that translates into a dissapointment over the whole GF100 architecture. So in general I'd say that those who aren't impressed by charts and graphs from the whitepaper should simply wait a month or two and they'll get their games performance numbers. For me getting GTX285+130% with MSAA 8x in HAWX is enough already not to be dissapointed with the provided info. As for the rest of performance numbers -- well, it's just a matter of time now.

I think someone already mentioned it, when comparing the GF200 and previous products (G92 etc) at 8X AA and saying the "new" generation product is X times faster, I don't think that is all THAT great when you look at how badly the GF200/G92 are at 8X AA.. the bar is set soo damn low that any % increase is going to look great. I do commend NV on their ability to finally match ATI's 8X AA performance, as the performance drop from 4X to 8X closely resembles what ATI has been doing now for a couple of generations.

 

[Groo The Wanderer]

My point is that we haven't been able to run Unigine ourselves yet on real hardware. Never in my entire time doing this have I ever used a pre-release benchmark from a hardware vendor to make a decision about real-world perf. Nor theoretical really. You you're calling it already for NV (even if you're eventually right, it pays to wait until you're 100% sure).

One thing to keep in mind, NV is only quoting a small portion of Unigine/Heaven for their numbers, and those are by far the most tessellation heavy portions of the benchmark. They are specifically not quoting the whole benchmark numbers. Think there is a reason for that? They don't even show numbers for heavy tessellation WITH other things going on. Hmm......

It's appears to be a pretty great graphics architecture from where I'm sitting, crying would be a bit silly. I'm not miffed you like it, it's how readily you (and others) do (and in the other direction with ATI hardware and its fanboys).

To me it appears to be a great GPGPU architecture with some graphics tacked on. I can't for the life of me figure out why a 2:1 SPP FP ratio is good for gaming, nor ECC capabilities, but then again, does that cost them anything other than die area, yield, and power?

Nope, but then I don't think anyone does outside of NVIDIA.

From what my sources are saying, you can safely extend that to many INSIDE Nvidia.

-Charlie

 

[Razor1]

well then we can always talk about synthethics where the the g100 will be upwards of 3 times + faster in shader synthetics in 3dmark tests, why didn't nV really talk about those, I have info that shows some tests are 5 times faster, come on they could have easily used to those if they wanted a pure PR stunt.

To answer some of your Q's about demos used, Heaven portion was heavy tesselation, they wanted to show you gaming DX 11 performance wasn't compromised because of the crazy notion of silicon being used for GPGPU.

Hawx, was used to show how the the new architecture can extend its lead over todays cards on today's games. Far Cry 2, older games still have performance increases.

Come on can't have everything and these arguements are pointless.

 

[Bob]

I like Charlie. First, he starts with the assumption that NVIDIA engineers don't know what they're doing. Therefore, anything that NVIDIA engineering does is bad. The latter is then used to demonstrate the initial premise.

The logic is perfect! A perfect circle, in fact.

 

[eastmen]

This sticks out like a sore thumb. The HD5850 for instance is cheaper than a GTX285 yet offers 20~30% more performance and you and AMD to lower the prices? amazing logic!

not for nothing but there is a pretty big gap between the msrps. Its $259 vs $379 or a $120 price diffrence.

I can see AMD being able to lower the price of the 5870 a bit. Perhaps down to $325 and make a profit.

 

[Periander]

So, to answer the question, yes, ATI can make a Hemlock for far cheaper than NV can make a GF100. In fact, they can price Hemlock at the point where NV is in the red and ATI is making money, albeit a very slim amount. If you add in margins for the AIBs, disti's, shipping companies, and (r)etailers, things get much uglier for NV.

What relevence do any of these calculations of margins, yields, die size ect. that you keep prattling on about have without including the professional side of things into the equation? The real bottom line is that Nv is generating a larger gross profit from Quadro alone than ATI is from its entirety.

 

[aaronspink]

Thanks to Rys, Trini and Jawed for some insight into memory usage differences and the reasons they may occur.

@Jawed - pure speculation here: I doubt they change the SM per GPC so soon unless it really is a "trivial" matter. Speculating because of the problems NV has had to market and the engineering hurdles they have encountered so far may have kept them too busy to re-engineer the parts with such a fine scalpel so soon unless the GF100 was inherently designed to be easily modular.

On second thoughts they did just go from 512 CUDA Cores to 448 CUDA Cores - I wonder what that entailed and what else was chopped.

The difference aligns with turning off 2 SMs either both in the same GPC or 1 in two different GPCs. But then again, we all know how *in*accurate some of nvidias descriptions and diagrams have been in the past...

For all we know the raster and setup units are right next to each other and output to a LL queue that the SM read from and the grouping they are showing doesn't really exist. Hopefully people still remember then trying to pass off x8 simd as 8 seperate cores, right?

 

[aaronspink]

Do you think video decoding improvements are interesting anymore as a USP? On the PC space do we need to decode more than 2 1080p streams in hardware or have we past the point of diminishing returns?

and lets be honest, if you are blowing 300W for the GPU in your system, you likely have a CPU that can easily decode 1080p. Only part of video that is at all interesting is encode, and until someone comes out with a GPU that runs x264, its not going to change much.

 

[aaronspink]

Maybe not 7x, and as we all know there is premium for top "Halo" performance parts. Part of the problem is there is no competition, meaning there is no reason for ATI to drop prices being that they are most likely selling all 5900 series parts at the price the market will bare.

And each 5970 they sell is 2 lost 5870 sales, so in a supply constrained situation, it only makes sense for them to sell the 5970 parts at 2x the price of the 5870 parts.

 

[willardjuice]

Let's try to focus on GF100 speculation and not speculation about Charlie. We already have created a separate thread for that! Last warning!

 

[trinibwoy]

One thing to keep in mind, NV is only quoting a small portion of Unigine/Heaven for their numbers, and those are by far the most tessellation heavy portions of the benchmark. They are specifically not quoting the whole benchmark numbers. Think there is a reason for that?

Yep, they wanted to highlight their advantage in tessellation heavy scenes. Dastardly, ain't it?

Could you expand a bit on what you consider "graphics" to be exactly since you seem to think that it has been tacked on to Fermi. What specifically about the 16 tessellators, 4 raster units, 64 texture units and 48 rops that are all fully integrated with Fermi's memory hierarchy and shader cores speaks to this notion of yours? On one hand you're grasping weakly at Fermi's high DP performance as some sort of black mark yet at the same time ignoring the fact that there is vastly more silicon devoted to pushing polys, improving IQ and all the other goodies that we like to see in our games, not the least of which are all the cool new things we're gonna get to see thanks to those darned compute capabilities.

 

[ZerazaX]

No, because i don't know how good GF100 will perform in games. But you can't compare Hemlock and GF100 in the unigine benchmark and come to the conclusion that nVidia wasted a lot of transistors in useless things.
This is a example of the microstuttering problem in the unigine benchmark with a 5970.

http://www.xtremesystems.org/forums/showthread.php?t=243190

You should take a look there, it debunks microstutter being a dual GPU only thing pretty nicely. When a single 5870 experiences the same oscillations, it's hardly what we think of microstutter...

Microstutter might be one of the most mis-understood, over-hyped, and un-important term we've seen in GPU land in the last few years

 

[Kakkoii]

To me it appears to be a great GPGPU architecture with some graphics tacked on. I can't for the life of me figure out why a 2:1 SPP FP ratio is good for gaming, nor ECC capabilities, but then again, does that cost them anything other than die area, yield, and power?

ATI added on ECC also. (although a useless version that can only detect errors and nothing else.)

 

[Bob]

EDC is not ECC.

 

[Squilliam]

According to the analyses I have seen, yes it does. If you go by raw silicon costs...

-Charlie

Regarding these raw silicon costs, it doesn't make sense to me that they would simply divide the # of dies by the wafer costs. Would they not realistically expect a higher proportion of the wafer costs be ascribed to the higher bin parts to keep their margins relatively stable throughout the range of bins?

In addition to this, with Hemlock (sorry Dave I did think about writing R800) they would also expect that the highest bin Cypress parts would also be ascribed the highest proportion of the per wafer costs?

I know at this point Frankenstein has a greater chance of a fully functional brain than a Fermi board but even so the higher bin chips are actually worth more so I don't understand why they aren't priced as such on an analysis.

 

[seahawk]

It is not the price they can ask for, but the price they have to pay for on average and how much they need to make money with a GF100. And as Charlie showed, NV is very unlikely to ever make money with a GF100, as it is not competitive compared to the Cypress / Hemlock and ATI can easily cut their prices so much, that NV will make a loss with every Fermi sold as a GeForce. because of that (and the numerous technical problems with the design), I would not expect to see many GF100 GeForce versions on the shelves till they reach something like 70% yield. So not before B2 or so, which means Seot or later.
All this teasing with Fermi whitepapers is just to show how desperate the situation is at NV. They are screwed and they know it. All they have is FUD to keep people from buying the superior and available ATI products.

Or some people just got fed wrong info by a not impartial source.

 

[rpg.314]

It is not the price they can ask for, but the price they have to pay for on average and how much they need to make money with a GF100. And as Charlie showed, NV is very unlikely to ever make money with a GF100, as it is not competitive compared to the Cypress / Hemlock and ATI can easily cut their prices so much, that NV will make a loss with every Fermi sold as a GeForce. because of that (and the numerous technical problems with the design), I would not expect to see many GF100 GeForce versions on the shelves till they reach something like 70% yield. So not before B2 or so, which means Seot or later.
All this teasing with Fermi whitepapers is just to show how desperate the situation is at NV. They are screwed and they know it. All they have is FUD to keep people from buying the superior and available ATI products.

Or some people just got fed wrong info by a not impartial source.

Do you know that GF100 will be sold in professional and HPC markets where the margins are sky high and AMD has next to 0% market share, despite having great consumer products?

GF100 will make a lot of money for nv as r&d for mainstream consumer market has been paid off already, and *profits* in quadro and tesla markets are worth a LOT.

 

[Robert Varga]

Booooring...

...Nv is generating a larger gross profit from Quadro alone than ATI is from its entirety.

Very interesting. Can you show us some real numbers?

Aaaand that benchmarks, performance analysis and other thoughts you're all posting here... Damn who do you think is interested?

Take it easy, it was just green goblins presentation for their best frineds. All cherry picked benchies.

And what about that video leaks... Do anybody here think it was real leak? ROFL... It's all just a game. Nvidias game with us, normal people, tech websites readers, potentioal buyers. "Can you read us [to my cam] that FPS numbers again?"

It was all just part of a plan, marketing. Every company is trying to sell their stuff, and this is preparation phase. Prepare people, let them speculate, then make a big boom release, then sell like hotcakes. Good marketing, nothing special.

Booooring... :smile:

 

[nagus]

2560x1600 is my idea of high resolution, 1920x1200~ish is something consoles from 2005/2006 pretend to render at.

here is the slide: http://hardocp.com/image.html?image=MTI2MzYwODIxNHh4VHN0ekRuc2RfMV81MF9sLmdpZg==

Now I can already see this turn into a semantics war, sigh "noes!, high resolution is 1680x1050" etc. For all we know NV could be talking about 1024x768 for that matter. It's their job, claim superb improvements, just don't tell what exactly improved and don't mention the circumstances... keeps the fans happy until they're really going to launch a part.

And indeed, if Fermi was faster than Cypress at 25x16x8AA, wouldn't we see claims about three, four , or five times AA improvement over GT200 at "high" resolution? It's not their habit to claim just about 2 times and be done with it, they need larger awesomebars.

http://www.hardware.fr/articles/770-23/dossier-amd-radeon-hd-5870-5850.html

you are absolutely right... this definitely puts the “up to 2x over GT200 @8xAA high res” in a different light now, does it? ;-)

Cypress is already almost twice as fast as GT200 - on average