NVIDIA GF100 & Friends speculation

[Man from Atlantis]

690/1380/3800 MHz​

Market release
12/8/2010

http://bbs.expreview.com/viewthread.php?tid=37829&from=recommend_f

 

[nyt]

690/1380/3800 MHz​

Market release
12/8/2010

http://bbs.expreview.com/viewthread.php?tid=37829&from=recommend_f

Ahah.. the real reason behind Cayman delay !!

 

[Xenus]

Do we know if it's 14 or 15 SM's?

 

[Sampsa]

690/1380/3800 MHz​

Market release
12/8/2010

http://bbs.expreview.com/viewthread.php?tid=37829&from=recommend_f

Looks like a photoshop from this one:

 

[Robin B]

Its not hard to see, the Nv logo on the fan is painted....... really bad work here.

 

[BRiT]

Ahah.. the real reason behind Cayman delay !!

Once again, No.

 

[caveman-jim]

Once again, No.

You keep forgetting to finish that sentence.

 

[onethreehill]

GeForce GTX 460SE Tested
http://www.hkepc.com/5782/page/3#view

 

[neliz]

http://en.expreview.com/2010/11/18/...er-consumptiontemperature-surfaced/12013.html

When cooler of GTX 480 was installed in GTX 580,GPU temperature reached critical temperature of 97℃ and it entered into temperature protection state.While GTX 580’s self-contained vapor chamber cooler was only with GPU temperature of 90℃,which was within safe range.

 

[AnarchX]

"GTX 590/595":
http://www.enet.com.cn/article/2010/1119/A20101119777146.shtml

Dual-GF110@384-Bit, quad display, 375W

 

[Sampsa]

"GTX 590/595":
http://www.enet.com.cn/article/2010/1119/A20101119777146.shtml

Dual-GF110@384-Bit, quad display, 375W

Judging from the pic one part of the text on GPU's heatspreader says: 1039A1. Week 39 2010. My GeForce GTX 580's GF110 says 1038A1.

 

[Kaotik]

Would nVidia really dare to go straight and release official card outside PCIE specs?

 

[Picao84]

Would nVidia really dare to go straight and release official card outside PCIE specs?

I dont think so. We are probably just seeing an engineering sample, and besides, correct me if im wrong, but the 375W number is derived from the Dual 8-pin, which does not have to mean that the card consumes as much. One thing to think about also, might be that, with the exception of GTX480, all other Fermi derivatives have some overclocking potential built in. So the 375W limit might be a treat to overclockers and not really mean that card will exceed 300W. Then again, it might be all marketing talk

 

[Kaotik]

I dont think so. We are probably just seeing an engineering sample, and besides, correct me if im wrong, but the 375W number is derived from the Dual 8-pin, which does not have to mean that the card consumes as much. One thing to think about also, might be that, with the exception of GTX480, all other Fermi derivatives have some overclocking potential built in. So the 375W limit might be a treat to overclockers and not really mean that card will exceed 300W. Then again, it might be all marketing talk

Just look at the 580 consumption, it's not that far off 480's, just few watts under in general, it's nowhere near enough for a product with 2 of those chips and TDP under 300W - not even with nV's new formula for TDP
(and I believe 2x8pin as themselves are already breaching the specs regardless of actual power draw)

 

[Picao84]

Just look at the 580 consumption, it's not that far off 480's, just few watts under in general, it's nowhere near enough for a product with 2 of those chips and TDP under 300W - not even with nV's new formula for TDP
(and I believe 2x8pin as themselves are already breaching the specs regardless of actual power draw)

Well, then look at Cypress and R5970... Cypress TDP is 188W which would break the 300W barrier if doubled. Same logic applies.

Remember that GTX580 has more cuda cores enabled than GTX480. I bet nVIDIA can do a Dual GF110, with both GPUs with near GTX470 specs. If GF110 has lower TDP than GTX480, how would a GF110 with 448 cores enabled fare? Probably under 200W, I guess.

Remember also that GTX295 is composed of two GTX275 chips, which have a TDP of 219W. However if this would be enough to beat Antilles.. I guess we will have to wait and see.

 

[Kaotik]

Well, then look at Cypress and R5970... Cypress TDP is 188W which would break the 300W barrier if doubled. Same logic applies.

Remember that GTX580 has more cuda cores enabled than GTX480. I bet nVIDIA can do a Dual GF110, with both GPUs with near GTX470 specs. If GF110 has lower TDP than GTX480, how would a GF110 with 448 cores enabled fare? Probably under 200W, I guess.

Remember also that GTX295 is composed of two GTX275 chips, which have a TDP of 219W. However if this would be enough to beat Antilles.. I guess we will have to wait and see.

Not everything has to be doubled, and in case of Cypress, the TDP to begin with was much smaller, if we throw TDPs aside, GTX580 consumes around 240-250W under heavy gaming load (Crysis Warhead), that's already 70-80W more than HD5870 does in the same test, so they have to cut a LOT more to fit within 300W even in just gaming cases

(numbers based on GTX480 and 5870 numbers in Crysis Warhead at XBit Labs, and substracting the average difference GTX480 and 580 (10-20W) have from 480 results as XBit doesn't have 580 numbers in yet)

 

[caveman-jim]

6 memory chips per gpu says it's not the top binned ASIC, like GTX 295 wasn't two 280/285 on a stick

 

[neliz]

maybe we should put GF100/110GX2 out of our mind and focus on GF104/114 GX2?

 

[Kaotik]

6 memory chips per gpu says it's not the top binned ASIC, like GTX 295 wasn't two 280/285 on a stick

6 per chip per side of the PCB perhaps?
Or 2gbit chips?

 

[Gipsel]

6 per chip per side of the PCB perhaps?

Probably.

Or 2gbit chips?

Are there any 64bit wide memory chips?

Nevertheless, the article linked by AnarchX also shows a GF100 based "480 X2" which has also never seen the light of the day. So how should one look at this other than as a design study?