The official GT300/Fermi Thread

[AbelJemka]

I doubt it'll be $600, NVidia said they realized they launched the GTX-280 too high and they won't be making that mistake again. I'm thinking probably around $400 at launch, which will put a VERY tight squeeze on ATi.

Of course, that's all theoretical as no one can know for sure how NVidia will price this thing. We'll find out soon enough though. Considering NVidia is aiming to market for tesla, where the profit margins are much higher, they can afford lower prices on the desktop cards and make the difference off tesla kits. Just a little something to keep in mind.

Quote 1 :

Jonah did step in to clarify. He believes that AMD's strategy simply boils down to targeting a different price point. He believes that the correct answer isn't to target a lower price point first, but rather build big chips efficiently. And build them so that you can scale to different sizes/configurations without having to redo a bunch of stuff. Putting on his marketing hat for a bit, Jonah said that NVIDIA is actively making investments in that direction. Perhaps Fermi will be different and it'll scale down to $199 and $299 price points with little effort? It seems doubtful, but we'll find out next year.

Quote 2 :

Last quarter the Tesla business unit made $10M. That's not a whole lot of money for a company that, at its peak, grossed $1B in a single quarter. NVIDIA believes that Fermi is when that will all change. To borrow a horrendously overused phrase, Fermi is the inflection point for NVIDIA's Tesla sales.

Quote 3 :

We'll see how that plays out, but if Fermi doesn't significantly increase Tesla revenues then we know that NVIDIA is in serious trouble.

[Wiggy McShades]

Quote 1 :


Quote 2 :


Quote 3 :

ya read that part about how they were NOT going to price the cards too high like gt200?

[gamervivek]

I doubt it'll be $600, NVidia said they realized they launched the GTX-280 too high and they won't be making that mistake again. I'm thinking probably around $400 at launch, which will put a VERY tight squeeze on ATi.

Of course, that's all theoretical as no one can know for sure how NVidia will price this thing. We'll find out soon enough though. Considering NVidia is aiming to market for tesla, where the profit margins are much higher, they can afford lower prices on the desktop cards and make the difference off tesla kits. Just a little something to keep in mind.

they priced it too high since they miscalculated the rv770's performance,this time round they might make sure that the performance is where the 650$ tag makes sense.Or at least the marketing hype does.

[ubuntu83]

they can afford lower prices on the desktop cards and make the difference off tesla kits. Just a little something to keep in mind.

They'll have to spend a lot on it's marketing. It's current market share is nothing to fancy about.

Last quarter the Tesla business unit made $10M. That's not a whole lot of money for a company that, at its peak, grossed $1B in a single quarter.

[AbelJemka]

they priced it too high since they miscalculated the rv770's performance,this time round they might make sure that the performance is where the 650$ tag makes sense.Or at least the marketing hype does.

I was about to post to answer Wiggy McShades but you burn me on it

I read this part like him.

Ujesh is wiling to take total blame for GT200. As manager of GeForce at the time, Ujesh admitted that he priced GT200 wrong. NVIDIA looked at RV670 (Radeon HD 3870) and extrapolated from that to predict what RV770's performance would be. Obviously, RV770 caught NVIDIA off guard and GT200 was priced much too high.

Ujesh doesn't believe NVIDIA will make the same mistake with Fermi.

[tajoh111]

Ok i give you what i thinking.

I will quote Anandtech article about Fermi :
http://anandtech.com/video/showdoc.aspx?i=3651&p=7


AMD launch it card before this time, what AMD is doing now : extrapolate GT300 performance and cost.
Performance? GTX285 SLI is like 30% faster than 5870 in average. GTX380 may be more like 50% to 60% faster than 5870 in average. Maybe even less.
Cost? 40% more transistors than RV870 and 384bits istead of 256bits. 600$? More?
Diltech speaks about GTX395 but in Nvidia history multigpu cards were launched very late (More than 6 months in average).
Basically AMD have 3 months to sell DX11 card with the help of Windows 7.

You can only extrapolate so much and do so much with the technology you have at the time.

AMD R and D budget is tiny compared to Intel and NV(especially Intel), you can over estimate your rivals performance by 1000% and it will do nothing if you don't have the r and d budget to get something going to match that estimate.

With so many losing quarters in the past(except a couple quarters lately), I can imagine AMD graphic division was working on a shoestring budget, especially when AMD itself is so in the hole. Thankfully the research ATI put into r600 before the AMD and ATI merger paid off to some extent with r7xx and possibly to an extent r8xx as it turned out r6xx turned out to be a very scalable architecture. However research for the next big thing I can imagine being lacking for AMD and if this thing performs 50-60% faster than rv870, then AMD will need to come out with something new and not just a bigger chip with more shaders as returns have started to decrease with more shaders.

It will take either a big chip from AMD(which seems to be against their design philosophy) or a new architecture. I think a new architecture is not coming any time soon because of budget issues.

What AMD did with the R8xx is stretch the limits of the design(which NV did with g80->g200) that began with r600, it's all you can do when your company doesn't have the money to design a new architecture.

[JohnJohn]

2x+ the speed of a GTX 285(not sli, literally double or more performance) will put a single GTX-380 far above the 5870(40 to 60%, if not more), and a simple overclock of the 5870(5890) is NOT going to catch it.

I concur with everything you said except this. If something has taught us the 5870, is that doubling everything up doesn't translate always in doubling performance. Even without having the SLI/CF handicap

[Chumbucket843]

I am excited to see exactly what they have done with the architecture but from what I have already seen they are dismissing their, original, primary mission of delivering uncompromising gaming performance and are now more focused on marketing and very minor niche markets. While there is much money to be made in the GPGPU market it is a little disappointing to see them chasing it so hard and with little to no regard to their gaming performance. Plus Nvidia's arrogance as a whole is very bewildering and insulting as a consumer.

i was like wtf when i saw 8x dp performance. i hope in the future they will have a version with less dp support. i dont use it so its just wasting power. the hpc market is 1/2 size of the desktop gpu market. it is clear they have a reason to target this market but i would have to agree this is overkill for a gpu and the money they spent on designing specifically for this market might not even work. i might end up buying larrabee or a 5870 if the tdp and price are not right. the videos on their site dont even mention gaming.

[LordEC911]

You can only extrapolate so much and do so much with the technology you have at the time.

AMD R and D budget is tiny compared to Intel and NV(especially Intel), you can over estimate your rivals performance by 1000% and it will do nothing if you don't have the r and d budget to get something going to match that estimate.

With so many losing quarters in the past(except a couple quarters lately), I can imagine AMD graphic division was working on a shoestring budget, especially when AMD itself is so in the hole. Thankfully the research ATI put into r600 before the AMD and ATI merger paid off to some extent with r7xx and possibly to an extent r8xx as it turned out r6xx turned out to be a very scalable architecture. However research for the next big thing I can imagine being lacking for AMD and if this thing performs 50-60% faster than rv870, then AMD will need to come out with something new and not just a bigger chip with more shaders as returns have started to decrease with more shaders.

It will take either a big chip from AMD(which seems to be against their design philosophy) or a new architecture. I think a new architecture is not coming any time soon because of budget issues.

What AMD did with the R8xx is stretch the limits of the design(which NV did with g80->g200) that began with r600, it's all you can do when your company doesn't have the money to design a new architecture.

Expect a shrink in '10 and a new architecture 1H '11, if not before.
You seem to not realize that architectures take 3-4years of design, not 2-3 months.

[v_rr]

Just asking if it's possible:

->GT300: Q1 2010


TSMC Will Move to 28nm Process in Early 2010
The company will be able to offer the 28nm process as a full node technology
http://news.softpedia.com/news/TSMC-...10-94593.shtml

Global Foundries:

32nm SOI technology will be shipping in 2010 in an AMD design though GF does have a 32nm bulk technology that they will begin accepting orders for in the second half of this year.

http://www.pcper.com/comments.php?nid=7237

32nm Bulk@ H2 2009

-> RV870@32/28nm H1 2010 (??)

RV870 will be alone on it's own league for some quite time. (3 months++)
So Fermi could face a new RV870 shrink node later?

[tajoh111]

Expect a shrink in '10 and a new architecture 1H '11, if not before.
You seem to not realize that architectures take 3-4years of design, not 2-3 months.

What the heck, I thought thats what my post implies. AMD won't be coming out with anything spectacular anytime soon because of the shoestring budget they have been working with because of so many bad quarters. AMD Graphic division has made a total profit of 50 million or so from its profitable quarters, but has lost hundred of millions if not billions in there quarters where they posted a loss.

AMD probably thought NV was going to go for a die shrinked G200 architecture as they had one in the pipeline until they decided to cancel it for gt300. As a result, I am not sure how much of an offence they can mount against gt300. All they can hope for is for gt300 to be a failure in efficiency, which NV I don't think will make because of the lessons learned from g200.

NV has been a much more profitable company overall and has probably been working on something pretty complex for the last 4 years as todays news confirms.

[ubuntu83]

What the heck, I thought thats what my post implies. AMD won't be coming out with anything spectacular anytime soon because of the shoestring budget they have been working with because of so many bad quarters. AMD Graphic division has made a total profit of 50 million or so from its profitable quarters, but has lost hundred of millions if not billions in there quarters where they posted a loss.

NV has been a much more profitable company overall and has probably been working on something pretty complex for the last 4 years as todays news confirms.

By the current specs of GT300 we know it's HPC performance will be really good but that can't be said about the gaming. Can you tell me how many of those 3 billion transistors are dedicated to 3D performance. I don't think it will be a ground breaking product as far as the 3D performance is concerned. Not any changes except for the MIMD units instead of SIMD.

[DilTech]

I concur with everything you said except this. If something has taught us the 5870, is that doubling everything up doesn't translate always in doubling performance. Even without having the SLI/CF handicap

The difference here is the GTX-380 is a brand new architecture, while the 5870 is still the same as the 4870 with double everything but the memory bandwidth and clock speeds. ATi are reaching the point of diminishing returns with the 5870, which is the problem we see and what stopped them from seeing double the performance. NVidia's new architectures have generally produced double the performance, if not more.

As for the fact that they aren't speaking about the gaming performance...of course not. They would tick off all their partners if they killed off their current gen sales by telling the consumer about their next gen performance.

[tajoh111]

By the current specs of GT300 we know it's HPC performance will be really good but that can't be said about the gaming. Can you tell me how many of those 3 billion transistors are dedicated to 3D performance. I don't think it will be a ground breaking product as far as the 3D performance is concerned. Not any changes except for the MIMD units instead of SIMD.

NV would be shooting itself in the foot, if they didn't offer good gaming performance as it is likely either company is going to have start to show their cards if they want their chip to be included in the next gaming console. 512 shaders with higher shader clocks should already increase performance significantly.

Even those the chip is massive, NV still seems to want more efficiency from it sharders. Additionally to what NV has said, the new shaders would have to be more powerful than the ones in the past if NV want this chip to scale past one generation. This should be obvious with such a large chip. So more powerful shaders + 512 seems to be a decent recipe for a card with good gaming performance.

The problem with AMD right now, is die shrinks are getting harder to come by as they sometimes take a while to successful implementate. Also as we have seen with the 5870, they are starting to hit a wall when it comes to adding more shaders = more performance. This can be seen with 5870 vs 5850 especially, as the 5850 is far more efficient gaming perfomance wise vs tflop than the 5870.

[ubuntu83]

The difference here is the GTX-380 is a brand new architecture, while the 5870 is still the same as the 4870 with double everything but the memory bandwidth and clock speeds. ATi are reaching the point of diminishing returns with the 5870, which is the problem we see and what stopped them from seeing double the performance. NVidia's new architectures have generally produced double the performance, if not more.

As for the fact that they aren't speaking about the gaming performance...of course not. They would tick off all their partners if they killed off their current gen sales by telling the consumer about their next gen performance.

What are the ground breaking changes that you think will bring exceptional performance for the 3D apps?

[tajoh111]

What are the ground breaking changes that you think will bring exceptional performance for the 3D apps?

This is kind of a guess, but I can imagine the new on chip cache reducing the hit AA has on the chip big time. Heck it might be even free up to a certain extent.

[gamervivek]

The difference here is the GTX-380 is a brand new architecture, while the 5870 is still the same as the 4870 with double everything but the memory bandwidth and clock speeds. ATi are reaching the point of diminishing returns with the 5870, which is the problem we see and what stopped them from seeing double the performance. NVidia's new architectures have generally produced double the performance, if not more.

As for the fact that they aren't speaking about the gaming performance...of course not. They would tick off all their partners if they killed off their current gen sales by telling the consumer about their next gen performance.

you could say the same for rv670-->rv770.It was 2.5 times but the performance was about 2x times,and iirc gtx280 was beaten by 8800gt SLI when it was introduced.Diminshing returns for the 5870 is too early to call for ati's arch,will they change it for something radically new depends on how far they want to take it up against nvidia in the gpgpu sector.

[Mechromancer]

It looks like Nvidia made a CPU that pretends to be a GPU. This thing will be a monster at GPGPU no doubt. There is no question about it; Fermi is a generation beyond AMD's GPGPU capabilities. The next generation of GPUs from both companies are truly exciting this round.

One has to wonder how this will affect OpenCL. GT300 can do DXCompute, OpenCL and C++. RV870 can only do DXCompute and OpenCL. Will C++/CUDA GPGPU programming take off and leave OpenCL in the dust? Or will OpenCL and DXCompute, as they're supported by both companies, reign supreme. I think it will come down to the flexibility in each implementation. C++ has a leg up on that front. Larrabee may be the deciding factor. 2010 will be an incredibly interesting year.

[zanzabar]

What are the ground breaking changes that you think will bring exceptional performance for the 3D apps?

im not sure that it will be better than 2 of the g200 in 3d, if it was ground breaking then we would have seen a bigger drop in price when the 5870 came out to remove all of the back stock or they will come dec
i want numbers

[XCheater]

If GT300 is a monster of general purpose computing, why don't we make the NVIDIA card do the general purposes computation, while ATI cards do the rendering?
While we used high-end multicore Intel/AMD CPU for "northbridge replacement"
It's a swap back to "northbridge" industry" but with much powerful substitution
That's might make standalone desktop computer a 10 TFlops machine BTW

[zerazax]

I don't see it being 2x G200 either. They keep saying new architecture, but from what I've seen, it's built heavily on G200, just like G200 was built heavily on G80. G200 was nearly 2x the units of G80, but it didn't hit 2x the performance until far after release when newer games were optimized/able to take advantage of what G200 had extra.

However, I think without any hard data on clocks, it's impossible to claim where it will end up. I'm hopeful it's good, but when I hear them admit that it's been delayed, that's usually not a positive sign

[Sh1tyMcGee]

Its so funny how people were counting nVidia out because they were late 1 week announcing their product. I purchased a 5870 because im so impatient. The nvidia Fermi is a advancement in GPU technology, where as ATI 5870 is just the same old thing just doubled. I am looking forward to purchasing the Fermi soon!

[jaredpace]

courtesy of bit-tech:

[xbrian88]

HwMania - Visualizza messaggio singolo - [Thread Ufficiale] nVIDIA Fermi

[RAW-Raptor22]

ISA???