NVIDIA GF100 & Friends speculation

[Kaotik]

Probably.
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?

The "480 X2" was IIRC "ASUS MARS II", not nV's own project

 

[Gipsel]

The "480 X2" was IIRC "ASUS MARS II", not nV's own project

May point was the answer to the question: Could you buy it?

 

[neliz]

The "480 X2" was IIRC "ASUS MARS II", not nV's own project

Different products, one is not mutually exclusive.

 

[caveman-jim]

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

What bus width are you imagining?

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

GF114.... all 8 SM's/TU's/ROP/Polymorphs enabled, bumped clocks, the microarchitecture improvemetns from GF110.

Makes the 465 useless, makes a case for new Quadro's/Tesla's, too.

 

[PSU-failure]

GF114.... all 8 SM's/TU's/ROP/Polymorphs enabled, bumped clocks, the microarchitecture improvemetns from GF110.

Makes the 465 useless, makes a case for new Quadro's/Tesla's, too.

I think 465 has ever been useless, so there have never been any need to render it useless.

On top of that, GF110 has ZERO improvement over GF104, so a "GF110-style improved GF104" would resemble GF104 very much.

 

[caveman-jim]

I think 465 has ever been useless, so there have never been any need to render it useless.

On top of that, GF110 has ZERO improvement over GF104, so a "GF110-style improved GF104" would resemble GF104 very much.

Pretty trolly post... can you support either statement with arguments or facts?

 

[Kaotik]

Pretty trolly post... can you support either statement with arguments or facts?

Well, it is true, apart from the 465-part perhaps, it was useful in sense of dumping as much GF100 chips out there as possible.

As for GF104-GF110, all the enhancements GF110 got over GF100 were already in GF104 (fullspeed FP16 bilinear and whatnot), in fact GF110 didn't even get all that was improved over GF100 in GF104 (video/media engine, missing HD audio bitstreaming support)
The claimed "APM" is at least at the moment nothing but marketing, it's nothing but driver based application recognition, it doesn't care if you OC the card to hell and back and go past the supposed "throttle values", in fact just using older version of Furmark is enough to bypass it and blast to those 350W consumptions.


I expect GF114 to be 0,99999999:1 identical to GF104, with perhaps some layout optimizations if anything.

 

[caveman-jim]

So the claimed microarchitecture improvements for performance and performance/watt in GF110 were really just already done in GF104?

 

[Unknown Soldier]

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

What about GF110GX2 on 28nm? Would 28nm be enough to lower the power requirement to reach the level specified?

IIRC TSMC did say that they would be producing 28nm by end of 2010.

28nm On Track with the Roadmap

Instead of 32nm fabrication process, TSMC concentrates on 28nm technologies. Initially the company plans to launch 28nm SiON/Poly process for low-power devices and later on it will roll-out high-k metal gate (HKMG) based technologies for more advanced products.

“The first node we are going to release for the 28nm will be called the 28LP. This is our poly gate and silicon oxide nitrate version. We will establish production at the end of June this year, about four months from now, and this is for the low power application. […] The first HKMG process we call 28HP for the high performance application will be introduced by the end of September this year, and followed by three months later December will be the 28HPL, which is [our] first high-k metal gate introduction for the low power application,” said Mr. Chiang.

http://www.xbitlabs.com/news/other/...to_Double_40nm_Output_by_End_of_the_Year.html

 

[PSU-failure]

Well, it is true, apart from the 465-part perhaps, it was useful in sense of dumping as much GF100 chips out there as possible.

Well... really, GTX465 has been useful only for that, since GTX460 has been introduced some 6 weeks later with better everything. Looking at how it performed (~GTX285), it wasn't even useful for anything else.

As for GF110, I find it dubious they would have removed full-speed FP16 while it was here with GT200 and back with GF104, power isn't that much lower either (considering some ~30-watt drop entirely due to the better HSF plus 50-watt higher worst-case power draw despite the supposed "improvements" in that area).

I already said it earlier: a "fully unleashed" GF104, or what's actually called GF114, would perform like AMD's "performance" high-end (HD6870) but draw as much as AMD's "enthusiast" low-end (HD6950) if not worse, with a cost similar to the second (as a cut-down version).

 

[Kaotik]

So the claimed microarchitecture improvements for performance and performance/watt in GF110 were really just already done in GF104?

That's from cutting those ~200m useless transistors & better cooling

 

[cal_guy]

As for GF110, I find it dubious they would have removed full-speed FP16 while it was here with GT200 and back with GF104, power isn't that much lower either (considering some ~30-watt drop entirely due to the better HSF plus 50-watt higher worst-case power draw despite the supposed "improvements" in that area).

GT200 did not have full speed FP16.

http://www.behardware.com/articles/723-5/product-review-the-nvidia-geforce-gtx-280-260.html

 

[rpg.314]

IIRC TSMC did say that they would be producing 28nm by end of 2010.

Yeah. Right.

 

[Triskaine]

IIRC TSMC did say that they would be producing 28nm by end of 2010.

TSMC also said that they would be producing 40 nm by the end of 2008.

 

[Kaotik]

Not sure wether this or review thread would have been better, but Palit is recalling their GTX580 Sonic -series cards
http://www.evga.com/forums/tm.aspx?m=694378

Dear customer

Thank you for purchasing Palit product.

We found few GPUs seem have unstable status in factory clock so we issue the recall notice.
Most of the GTX 580 Sonic products don't have such an issue but we rather to have more conservative consideration so we issue the recall notice.
Please contact your vendor to proceed further recall steps.
Thanks.

Palit Support

 

[no-X]

So the claimed microarchitecture improvements for performance and performance/watt in GF110 were really just already done in GF104?

If you switch-off the furmark-protection and use GTX480-cooler, GTX580 will consume more power, will reach higher temperatures and higher noise, than GTX480. So, these "improvements" are quite debatable, at least when speaking about GPU-level.

 

[Sxotty]

If you switch-off the furmark-protection and use GTX480-cooler, GTX580 will consume more power, will reach higher temperatures and higher noise, than GTX480. So, these "improvements" are quite debatable, at least when speaking about GPU-level.

That is not correct. The results I saw showed them consuming the same power. And the GTX580 still gets more performance/watt in that case.

 

[no-X]

With the same cooler?

 

[neliz]

With the same cooler?

Good question, I haven't seen power consumption with the 480/580 cooler swap, only the temps and they are equal for the 580/480.

http://en.expreview.com/2010/11/18/...-consumptiontemperature-surfaced/12013.html/6

 

[PSU-failure]

Good question, I haven't seen power consumption with the 480/580 cooler swap, only the temps and they are equal for the 580/480.

http://en.expreview.com/2010/11/18/...-consumptiontemperature-surfaced/12013.html/6

The temps you show are far from "equal"... 97°C and throttling vs 90°C.

BTW, Expreview isn't what I'd call a reference.