The real die sizes for gf100 and gf104. Surprize for ya.

[happy medium]

The full gf 104 A.K.A. gtx 475 is actually smaller then the 5870 Cypress chip.
The gtx 480 is smaller then the gtx 285 series.
Other good info in the article.

http://www.brightsideofnews.com/news/2010/8/9/nvidia-fermi-geforce-die-sizes-exposed.aspx

Dupe
-ViRGE

 

[busydude]

I guess this has been discussed few days back.. and I am pretty sure you were one of the posters there.

 

[happy medium]

I guess this has been discussed few days back.. and I am pretty sure you were one of the posters there.

really?, if so I forgot.
Edit;
I went back 6 days. I didn't see this article.

 

[jvroig]

The full gf 104 A.K.A. gtx 475 is actually smaller then the 5870 Cypress chip.

I have no trouble accepting this, given that the GTX 460 is a great balance of heat, power, and performance. I can't get a better AMD card anywhere near its price range: 5770 is cheaper but far less powerful to realistically compete, and the 5830 is some sort of bizzaro-bastard-card whereas it has the "Radeon" branding but inherited Fermi's worse characteristics - I'm not touching that with a 10-foot pole, not even with a 10.5 foot pole while wearing gloves, for that matter.

The gtx 480 is smaller then the gtx 285 series.

Congrats to Fermi GF100, I suppose, for being smaller than the gtx 285 (whatever that may mean). I know this won't make the GTX 480/470 cards suddenly less hot or less power hungry or anything, but hey, at least it's not bigger than the gtx 285, for whatever that's worth (I can't imagine it's worth anything at all - maybe to debunk some old Charlie facts/statements?)

 

[GaiaHunter]

So basically they are saying that GF100 is 530 mm^2 which has been known for what, 1 year?

About the GF104 we have BSN saying one thing (without any reasoning for it) and Charlie and Nordich hardware saying another thing - http://www.nordichardware.com/news/71-graphics/40710-geforce-gtx-460-core-revealed.html.

Considering this has already been discussed on these forums, I don't see where is the surprise.

 

[RussianSensation]

Considering this has already been discussed on these forums, I don't see where is the surprise.

I believe most websites have initially incorrectly stated that GTX460 (GF104) is actually larger in size than Cypress chip is. It looks like NV trimmed a lot of 'fat' from GF100 to get GTX460 performance in a package that's 200mm2 smaller! That's an amazing accomplishment considering GTX460 is about 20% slower than GTX470 but GF100 is a 60% larger die than GF104!

The reference about Fermi being smaller than GT200, and actually not being much larger than original G80, helps to highlight the fact that NV's strategy has always been manufacturing large die GPUs. As far as NV's strategy is concerned, Fermi was no different than last 2 very successful generation of cards. So the consequence of poor execution not only rests with NV but TSMC since they had no problems manufacturing larger dies before on older processes.

If Fermi's pricing was not an issue at launch, no one would be talking about its size. Afterall, G80s (8800GTX, 8800GTS320/640) and GTX260/275/280/285 sold extremely well compared to ATI cards at the time. Granted, ATI destroys NV in terms of efficiency and performance / watt :awe: However, if we start to look at performance in majority of DX11 games, it becomes pretty obvious that Tessellation engines/geometry, at least in part, contributed to Fermi's bloated die. It will be interested to see how large ATI's 6000 chip will be once it incorporates improvement in Tessellation performance.

 

[happy medium]

So basically they are saying that GF100 is 530 mm^2 which has been known for what, 1 year?

About the GF104 we have BSN saying one thing (without any reasoning for it) and Charlie and Nordich hardware saying another thing - http://www.nordichardware.com/news/71-graphics/40710-geforce-gtx-460-core-revealed.html.

Considering this has already been discussed on these forums, I don't see where is the surprise.

I think it is surprising that the gtx 475 is smaller then the 5870 but should give more performance. (I think its a good guess).
What a turn around for Nvidia is what I'm saying.

They went from too big, too power hungry, too hot, to smaller ,faster, as cool and using a little more power. And they did it without a die shrink for the gf100.

 

[evolucion8]

Granted, ATI destroys NV in terms of efficiency and performance / watt :awe:

I strongly agree with you, seems that ATi's die is doing more work per mm2 than nVidia's counterpart. The GTX 460 might be .04mm2 smaller, but is also slower than the HD 5870.

 

[Cookie Monster]

I strongly agree with you, seems that ATi's die is doing more work per mm2 than nVidia's counterpart. The GTX 460 might be .04mm2 smaller, but is also slower than the HD 5870.

That remains to be seen. A full fledged GF104 core (potentially the GTX475) could be as fast or faster than the full fledged cypress chip (HD5870 in this case).

The way I see it is that the cypress chip is alot like G70 from nVIDIA. It wasn't a radical architectural changed chip that looked at the long term performance but rather short term performance based on the market demand and the level of technology (in software/hardware) at the time. They manage to produce a chip that is quite impressive in the performance/watt or mm^2 area and managed to come to the market in a timely fashion because of this.

But the problem here is that nVIDIA has the better foundations than AMD in terms of GPU architecture. nVIDIA has made the big leap already and can afford to now optimize an architecture that has a lot of potential (in tessellation, DX11 performance, GPGPU etc). However AMD is yet to make this move and undoubtedly whatever they release next will decrease the perf/mm^2 gap advantage they had against nVIDIA.

 

[busydude]

really?, if so I forgot.
Edit;
I went back 6 days. I didn't see this article.

It was 10days back, exact same day the article went live. You were the last poster.

http://forums.anandtech.com/showthread.php?t=329882

 

[Grooveriding]

That remains to be seen. A full fledged GF104 core (potentially the GTX475) could be as fast or faster than the full fledged cypress chip (HD5870 in this case).

The way I see it is that the cypress chip is alot like G70 from nVIDIA. It wasn't a radical architectural changed chip that looked at the long term performance but rather short term performance based on the market demand and the level of technology (in software/hardware) at the time. They manage to produce a chip that is quite impressive in the performance/watt or mm^2 area and managed to come to the market in a timely fashion because of this.

But the problem here is that nVIDIA has the better foundations than AMD in terms of GPU architecture. nVIDIA has made the big leap already and can afford to now optimize an architecture that has a lot of potential (in tessellation, DX11 performance, GPGPU etc). However AMD is yet to make this move and undoubtedly whatever they release next will decrease the perf/mm^2 gap advantage they had against nVIDIA.

How much of a difference do you think another 48 shaders and a clock speed bump will give it ?

How much of a difference was there between the 260 and the 260 216core card, it was fairly minimal as I recall, 5% or so.

I reckon they will release a fully unlocked gf104 with a clock speed bump. The 460 needs about a 25% performance boost to match a 5870, will it get that from the extra shaders and clock speed increase ? Who knows, could be 15%, 20%, 25%, 30%. It will definitely bring it close.

How much will it cost ? Probably $300. I think the big thing nvidia wants to do with them is release a dual-gpu card so they can reclaim the flagship, they've been without it for an entire year now, that's a while for nvidia to have not had the fastest video card, longest I can remember.

Either way, 460 is probably not the future for them, as has been said earlier, it had to have been in the pipe for development alongside the gf100, to release at the mid-range.

I'm interested to see if going forward they continue with the trimmed down style of the gf104 for their flagships, or refine on the gf100 with all its additional features and continue to offer a chip with those abilities to the consumer market. I think it may be the latter with all the talk they give to gpu compute and what it can offer.

 

[jvroig]

It was 10days back, exact same day the article went live. You were the last poster. http://forums.anandtech.com/showthread.php?t=329882

What do you know, this is a repost after all, same link posted by the OP.

But nobody bothered to chime in there, maybe the OP wasn't too happy about it and decided this deserved more discussion as a thread of its own.

 

[Fayd]

I have no trouble accepting this, given that the GTX 460 is a great balance of heat, power, and performance. I can't get a better AMD card anywhere near its price range: 5770 is cheaper but far less powerful to realistically compete, and the 5830 is some sort of bizzaro-bastard-card whereas it has the "Radeon" branding but inherited Fermi's worse characteristics - I'm not touching that with a 10-foot pole, not even with a 10.5 foot pole while wearing gloves, for that matter.


Congrats to Fermi GF100, I suppose, for being smaller than the gtx 285 (whatever that may mean). I know this won't make the GTX 480/470 cards suddenly less hot or less power hungry or anything, but hey, at least it's not bigger than the gtx 285, for whatever that's worth (I can't imagine it's worth anything at all - maybe to debunk some old Charlie facts/statements?)

in a strict sense, it means they can get more cores per silicon wafer making them cheaper to produce. (at least, that 1 aspect of their production is cheaper. regarding the entire process, who knows. i'm sure more variables impact it.)

 

[Wreckage]

I really doubt anyone gives two bits about the size of their GPU.

Once you spend more than $50 on a video card, all that matters is how fast and what features it brings to games.

 

[evolucion8]

It was 10days back, exact same day the article went live. You were the last poster.

http://forums.anandtech.com/showthread.php?t=329882

Look at the nVidia threads that he started, is astonishing.

How much of a difference do you think another 48 shaders and a clock speed bump will give it ?

How much of a difference was there between the 260 and the 260 216core card, it was fairly minimal as I recall, 5% or so.

I reckon they will release a fully unlocked gf104 with a clock speed bump. The 460 needs about a 25% performance boost to match a 5870, will it get that from the extra shaders and clock speed increase ? Who knows, could be 15%, 20%, 25%, 30%. It will definitely bring it close.

How much will it cost ? Probably $300. I think the big thing nvidia wants to do with them is release a dual-gpu card so they can reclaim the flagship, they've been without it for an entire year now, that's a while for nvidia to have not had the fastest video card, longest I can remember.

Either way, 460 is probably not the future for them, as has been said earlier, it had to have been in the pipe for development alongside the gf100, to release at the mid-range.

I'm interested to see if going forward they continue with the trimmed down style of the gf104 for their flagships, or refine on the gf100 with all its additional features and continue to offer a chip with those abilities to the consumer market. I think it may be the latter with all the talk they give to gpu compute and what it can offer.

You are totally right, look at the full 512SP GTX 480, with more 32SP, it was 5.6 percent average faster across all scenarios, so if we account for 48 more shaders for the GF104, the estimate performance should be about 7.2 percent faster, which is not enough to beat a single HD 5870. So a full fledge GF104 at the same clocks as their single GPU counterparts should be enough to beat slightly the HD 5970 thanks to better GPU scaling, but considering its power consumption which is less than stellar compared to Cypress, I really doubt it.

 

[Wreckage]

You are totally right, look at the full 512SP GTX 480

Has this card actually been released or are you portraying speculation as fact?

 

[Cookie Monster]

How much of a difference do you think another 48 shaders and a clock speed bump will give it ?

How much of a difference was there between the 260 and the 260 216core card, it was fairly minimal as I recall, 5% or so.

I reckon they will release a fully unlocked gf104 with a clock speed bump. The 460 needs about a 25% performance boost to match a 5870, will it get that from the extra shaders and clock speed increase ? Who knows, could be 15%, 20%, 25%, 30%. It will definitely bring it close.

Well it would have another 48 shaders, an extra ploymoprh engine and extra 8 TMUs. The extra ALU increase alone is 14 percent over the GTX460. Given the appropriate clock frequencies e.g 750/1500MHz, we could theoretically match the GTX470 with less power consumption to its replacement.

How much will it cost ? Probably $300. I think the big thing nvidia wants to do with them is release a dual-gpu card so they can reclaim the flagship, they've been without it for an entire year now, that's a while for nvidia to have not had the fastest video card, longest I can remember.

Either way, 460 is probably not the future for them, as has been said earlier, it had to have been in the pipe for development alongside the gf100, to release at the mid-range.

Price would be dependent on the market situation/demand and other variables such as GTX465/470 inventory. But one thing for sure is that its probably cheaper to produce this hypothetical GTX475 than the GTX470.

I'm interested to see if going forward they continue with the trimmed down style of the gf104 for their flagships, or refine on the gf100 with all its additional features and continue to offer a chip with those abilities to the consumer market. I think it may be the latter with all the talk they give to gpu compute and what it can offer.

When we compare the GF104 and GF100, they are essentially the same architecture but with a changed emphasis on certain parts of the chip. GF100 was a little to ambitious in the sense that it tried to do it all. GF104 on the other hand cuts down some of the focus of GPGPU related functionalities/performance and focuses on gaming performance by adjusting the ALU/TMU ratio. (Although there are evident bottlenecks because of the reconfiguration, im not sure how these will turn out in the long term)

But a refined GF100 on a new silicon revision is well within the range of possibilities seeing as 28nm is not around til sometime next year. Id like a single chip refresh but because of SLi scaling and the performance of GTX460s, a dual GF104 SKU for the high end doesn't sound too bad either.

 

[evolucion8]

Has this card actually been released or are you portraying speculation as fact?

Here's the link to clear your mind and your way to portray speculation as facts, thank you!!

http://en.expreview.com/2010/08/07/more-benchmarking-results-of-512sp-gtx-480-exposed/9018.html

http://www.brightsideofnews.com/news/2010/8/13/512-core-geforce-gtx-480-reviewed-any-surprises.aspx

 

[happy medium]

Look at the nVidia threads that he started, is astonishing

In my defense, can you name one news worthy ATI acticle in the last 6 months? Besides S.I. speculation and the fact that they need to lower prices?
Really now, they haven't released a card in over 9 months.

Look at the first and second pages of the video forums, how many times do you see the words "gtx 460". There is a reason for that, it's newer technology.

I strongly agree with you, seems that ATi's die is doing more work per mm2 than nVidia's counterpart. The GTX 460 might be .04mm2 smaller, but is also slower than the HD 5870.

The full gf104 vs the 5870 will tell the whole story. It could wind up being smaller and faster then Cypress, that was my point.

edit:
And I did forget I posted this 10 days ago at the end of someone elses thread.

 

[Will Robinson]

Does a day pass without another "OMG NVidia is the shit,I'm so thrilled" post here?
Does NV just unleash shillbots via a timer or something?

 

[Keysplayr]

Umm, Will?

I haven't seen you start any threads that are exactly "pro NV" either.
Granted, there isn't anywhere near as many, but the ones you did start are anti-NV only.

So, before you go casting stones, you have to move out of the glass house.

Now, why doesn't everybody just take a step back and stop stabbing at each other?