Web-Site Displays GeForce GTX 480 Graphics Card with 512 Stream Processors.

[Skurge]

http://en.expreview.com/2010/08/04/...-testing-of-geforce-gtx-480-preview/8878.html

http://www.xbitlabs.com/news/video/...Graphics_Card_with_512_Stream_Processors.html

This I would like to see come out.

 

[lsv]

Fake?

 

[RussianSensation]

Probably stock piling these to counter 6870. The problem is GTX480 runs pretty hot. So a 512 SP version will consume even more power. I would have much rather seen a 384 SP GF104 clocked at 850-925 mhz to replace the GTX470.

 

[Sylvanas]

I think I'll hold out a little longer for an upgrade. With 6.25 % more execution resources and probably higher clockspeeds I wouldn't be surprised if it turns out being 10% faster than a 480 in ideal circumstances.

 

[OCGuy]

Probably stock piling these to counter 6870. The problem is GTX480 runs pretty hot. So a 512 SP version will consume even more power. I would have much rather seen a 384 SP GF104 clocked at 850-925 mhz to replace the GTX470.

As an owner of a GTX480, I am sure that a fully functional GF100 will consume it's share of power.

However, I think you are basing your assumption off the immature version that nV was forced to rush out to stop the bleeding.

"6870", if still on the current node, is only a half-gen that AMD is being forced to put out due to TSMC's issues.



The only reason I doubt the article is because nV would not keep the same name...it would be a 485 or "485 512" in my opinion.

 

[Sylvanas]

The only reason I doubt the article is because nV would not keep the same name...it would be a 485 or "485 512" in my opinion.

No doubt it will change. Early samples that appear in these 'leaks' are usually not detected correctly by software (GPU-Z / CPU-Z - unrecognised Device ID strings) or are running an immature BIOS revision (which in this case is probably an extension of their 480 BIOS).

 

[SHAQ]

Two 8 pin plugs? I wonder if they will upgrade the cooling for this 300 watt card.

 

[lopri]

B0 stepping?

 

[RussianSensation]

Honestly, for the few who are cruising at 2560x1600, with Crysis 2 delayed until next year, there is no need for anything faster in 2010 from what's currently available.

 

[Skurge]

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

Crysis bench.

Not much difference really.

 

[happy medium]

Two 8 pin plugs? I wonder if they will upgrade the cooling for this 300 watt card.

I think this is good news. I have this feeling these will overclock really well.
I read somewhere that these will possibly use less power at stock then a gtx480 with 480 sp's due to process refinements.

Funny thing is, they also said it will launch at higher clocks, but I don't see this in Skurge's link?
Pixel Fillrate/Texture Fillrate are still remain the same as 480core? Fake?

 

[edplayer]

Probably stock piling these to counter 6870. The problem is GTX480 runs pretty hot. So a 512 SP version will consume even more power.

Yes, but they should be able to find customers for them because there are always some that want the best (even if they are power hogs).

I would have much rather seen a 384 SP GF104 clocked at 850-925 mhz to replace the GTX470.

Nothing is stopping them from doing both!

However, I think you are basing your assumption off the immature version that nV was forced to rush out to stop the bleeding.

I doubt that Nvidia has made any significant revisions to the GF100 since its launched. Don't they write stepping codes on their gpus. Was the GF100 launch product using A3? If they had newer steppings, we would likely see some cards using them already for sale (even if they aren't 512 shader enabled).

 

[happy medium]

I doubt that Nvidia has made any significant revisions to the GF100 since its launched. Don't they write stepping codes on their gpus. Was the GF100 launch product using A3? If they had newer steppings, we would likely see some cards using them already for sale (even if they aren't 512 shader enabled).

No, it should be the B0 stepping.

 

[edplayer]

any photos showing that or just speculation?

 

[happy medium]

any photos showing that or just speculation?

Speculation, mixed with a little common sense.
If they couldn't get it right with the a3 silicon,what makes you think they didn't try again? I believe the next step is the B0 stepping.

 

[edplayer]

They probably tried a lot of times (way more than 3). Makes sense that the next stepping will have a different name but it could be like RussianSensation suggested, that they were yielding 512 shader parts for a while already and they were being stockpiled.

 

[happy medium]

They probably tried a lot of times (way more than 3). Makes sense that the next stepping will have a different name but it could be like RussianSensation suggested, that they were yielding 512 shader parts for a while already and they were being stockpiled.

I was thinking more along the lines of a gtx280 to gtx285 without the shrink.
A respin could get them better thermals ,higher clocks and less leakage.
A much better all around chip.

Wishfull thinking probrobly, but definitly possible.
Me I think the gtx 475 is gonna be the much better chip,just not as fast.
Its fun to speculate though,this is a specultion thread.

 

[Keysplayr]

Bogus. Texture fill rates are the same in both screenshots. Shouldn't the 512 sp 64 texture unit increase texture fill rates?

 

[happy medium]

Bogus. Texture fill rates are the same in both screenshots. Shouldn't the 512 sp 64 texture unit increase texture fill rates?

yea, I pointed that out, I think fake also.

quote:
"Funny thing is, they also said it will launch at higher clocks, but I don't see this in Skurge's link?
Pixel Fillrate/Texture Fillrate are still remain the same as 480core? Fake? "

 

[edplayer]

I was thinking more along the lines of a gtx280 to gtx285 without the shrink.
A respin could get them better thermals ,higher clocks and less leakage.
A much better all around cheap.

Wishfull thinking probrobly, but definitly possible.

How many times have they tried already?

I don't know but I would guess, computer simulation, at least 100 times. Test wafers, maybe 25 at least?

I know very little about semiconductor manufacture so this is just my guess. We know that they will test for speed, power consumption and yield but there are probably lots of other things they check that we common folk don't know about. I'm thinking that if they had better than A3 silicon product (now), it would likely be for sale in some of the current GTX 480.

 

[happy medium]

I'm thinking that if they had better than A3 silicon product, it would likely be for sale in some of the current GTX 480.
__________________

I'm thinking by now they would have released it allready (in small quanities like the 5970) if it was possible?
I'm thinking mabe they needed to respin to get the performance they wanted.
Why release it if it isn't at least 15 percent faster then the gtx480. This is why I believe it will also sport higher clocks with more sp's and be more overclockable.

 

[Sylvanas]

Bogus. Texture fill rates are the same in both screenshots. Shouldn't the 512 sp 64 texture unit increase texture fill rates?

It does. Probably unsupported by GPU-Z. Look at the results of the Vantage feature test 1 (filtrate). The 480 scores 38.82 GTEXELS/S and the 512 scores 41.55 GTEXELS/S. Sooner this is out the better, I'm itching for an upgrade.

 

[happy medium]

It does. Probably unsupported by GPU-Z. Look at the results of the Vantage feature test 1 (filtrate). The 480 scores 38.82 GTEXELS/S and the 512 scores 41.55 GTEXELS/S. Sooner this is out the better, I'm itching for an upgrade.

If it is real ,say goodby to the gtx 465,470 and 480. They are gonna be replaced by the gtx460, 475, and 485. Which is't exactly a bad thing.

 

[edplayer]

I'm thinking mabe they needed to respin to get the performance they wanted.

What have they been doing the past 18 months? I'm sure they have tried many possibilities already.

I'm not saying that it can't be a new stepping, it very well might be. I'm saying there is nothing that guarantees that it has to be. It could have been in production for a few months already but with very low yields.

Why release it if it isn't at least 15 percent faster then the gtx480. This is why I believe it will also sport higher clocks with more sp's and be more overclockable.

If there were 512 shader parts on A3 silicon, there is no reason to throw them away, even if it is only 14% (or less) faster. Even if all of the parts only added up to $500,000, I'm sure Nvidia would still choose to sell them.

 

[Sylvanas]

They now have a full blown review of the 512SP Fermi. Clocks are now 801/1601/950 so a serious bump there. Also look at that exotic cooling solution- I doubt that will make it to final production, with out of this world power consumption, 2x8pin PCI-E connectors and that cooling- this looks to me more like a prototype custom card by a partner and not a reference design. Performance is inline with a 6.25% increase in execution resources at the same clocks.