The GT300/Fermi Thread

[ajaidev]

Yep thanks hobbes

[neliz]

Even AMD already calculated theoretical performance of fermi just based on it's specs, they already made a comparison chart between r800 and fermi. I'm not asking for theoretical performance, anyone can do that, i was asking for concrete numbers and if the nvidia engineers can't do the math by now, based on the results of the running cards, Nvidia you're done, you're engineers are morons.

Of Course NV knows it's (theoretical) performance, otherwise they've been grabbing the numbers they're giving to partners/press over the past couple of months straight out of their asses.

[JohnZS]

Of Course NV knows it's (theoretical) performance, otherwise they've been grabbing the numbers they're giving to partners/press over the past couple of months straight out of their asses.

Joking aside, 20% faster for the Fermi which we will see in March sounds good... especially as we know that there is a more powerful variant coming in Q3. What are the odds we see the 360 6months before the 380?

As in March for the 448 piper (20% faster than HD 5870) and September for the 512piper which would trade blows with the HD5970?

Ati will most likely release a HD 5890 come May which will significantly narrow the Fermi performance lead to just a few %.

If ATi can get a 2GB at least 950Mhz or even 1Ghz HD 5890 out of the door it will be the first time I have purchased an ATi card since the Radeon 9700 Pro. If not then I will most likely stick it out with a working GTX 285 OCX I receive from RMA until something with more VRAM comes along.

John

[neliz]

It seems both cut-down and full part will be launching in March, placed on top of the GTX275 in the price list. Dual Fermi a couple of weeks after that. can't say if the cut-down part is 384 or 448. TDP for the "big" part should at least match HD5970.

[JohnZS]

It seems both cut-down and full part will be launching in March, placed on top of the GTX275 in the price list. Dual Fermi a couple of weeks after that. can't say if the cut-down part is 384 or 448. TDP for the "big" part should at least match HD5970.

Really?
Hmm nice. So, what IS coming in Q3? (other than the Daddy of all Quadro cards).
Could this be an Ultra Part?
What everyone needs to remember is that the ASUS M.A.R.S was essentially a rebaged Quadro FX 5800

Whatever is coming in September, is going to be huge (6GB of VRAM for a start). Could this be the dual Fermi?
Perhaps the cut-down part has 1.5GB of VRAM, the full part 3GB of RAM and the Dual part "6" GB?

It would make sense as OpenCL, DirectCompute, CUDA etc love the VRAM too you know..
John

[annihilat0r]

It seems both cut-down and full part will be launching in March, placed on top of the GTX275 in the price list. Dual Fermi a couple of weeks after that. can't say if the cut-down part is 384 or 448. TDP for the "big" part should at least match HD5970.

HD5970's TDP is 300w if I am not mistaken, so as far as I know there can't be a "at least as high as hd5970" situation, there's no higher than 300w in pcie tdp... plus, how is dual Fermi gonna be made with such TDP's with the single version, I don't know.

[zalbard]

5970's TDP is 294W iirc.

[Teemax]

HD5970's TDP is 300w if I am not mistaken, so as far as I know there can't be a "at least as high as hd5970" situation, there's no higher than 300w in pcie tdp... plus, how is dual Fermi gonna be made with such TDP's with the single version, I don't know.

I believe a dual Fermi would come soon. One of the reasons why NVIDIA has managed to built quite a number of loyal followers in the past years was their "fastest card on the market" claim. I don't think NVIDIA would want to lose that claim, especially after all the heat Fermi has taken.


For a dual Fermi, they could especially bin for chips that can undervolt, lower the clocks and/or disable some cores (like they did with GTX295). Single Fermi needs to push the power envelop to reclaim the "fastest GPU" title; but NVIDIA will manage a dual Fermi as well, IMHO.

[570091D]

you might find much truth in that Charlie article that showed G92 and GT200b life well into 2010.

i'm still running g92's, they've served me well, it's a good product when priced right. i just want to see at least 3 gf100 based geforce cards.

Even AMD already calculated theoretical performance of fermi just based on it's specs, they already made a comparison chart between r800 and fermi. I'm not asking for theoretical performance, anyone can do that, i was asking for concrete numbers and if the nvidia engineers can't do the math by now, based on the results of the running cards, Nvidia you're done, you're engineers are morons.

Good point on the drivers though, like i said before, didn't know NDA lifts next week, just wondering about another demo of running cards without numbers. Let's just wait then.

i agree with you, we can wait, true performance is really a year away with driver maturity....

btw, that ati marketing chart is probably the worst peice of negative pr i've EVER seen....

[Andrea deluxe]

http://forum.beyond3d.com/showpost.p...postcount=3366

http://forum.beyond3d.com/showthread...84#post1377784

[zalbard]

GTX395 is 60% faster than GTX380
GTX395 is 70% faster than HD5970
Release Date : May 2010, Price : 499-549 USD.

Yeah, right...

[annihilat0r]

I believe a dual Fermi would come soon. One of the reasons why NVIDIA has managed to built quite a number of loyal followers in the past years was their "fastest card on the market" claim. I don't think NVIDIA would want to lose that claim, especially after all the heat Fermi has taken.


For a dual Fermi, they could especially bin for chips that can undervolt, lower the clocks and/or disable some cores (like they did with GTX295). Single Fermi needs to push the power envelop to reclaim the "fastest GPU" title; but NVIDIA will manage a dual Fermi as well, IMHO.

I'm talking based upon the points neliz made. He said that a single Fermi's TDP would be 300w. Now you could take a stripped Fermi part and downclock it even further maybe like you said to make a dual Fermi, but it would still have a TDP of 300w and it would be SLI.

So a dual part drawing the same power as the single GPU part, and with SLI would probably mean being actually slower than the single GPU part. I don't get how a dual Fermi is going to be made if a single Fermi draws 300w

[annihilat0r]

Yeah, right...

LOL indeed. 70% faster than a 5970 at the same power level? $500?

All the while GTX 380 is faster than a HD5970 (because of the 60% 70% thing)

Right.

[RPGWiZaRD]

499-$549? FAIL, also 640-bit? Man if it could pack all those things listed at 40nm then this card would have to be like probably at least 35~40cm long which I doubt will happen.

[zalbard]

499-$549? FAIL, also 640-bit? Man if it could pack all those things listed at 40nm process then this card would have to be like probably 40cm long which I doubt will happen.

It could be as tall as MSI 5870 Lightning!

[RPGWiZaRD]

It could be as tall as MSI 5870 Lightning!

Good point there!

[LinkinParkBoy]

i think this is what you are talking about:

from anandtech
"The demo also used NVIDIA's sterescopic 3D technology - 3D Vision. We're hearing that the rumors of a March release are accurate, but despite the delay Fermi is supposed to be very competitive (at least 20% faster than 5870?). The GeForce GTX 265 and 275 will stick around for the first half of the year as Fermi isn't expected to reach such low price/high volume at the start of its life."

[neliz]

plus, how is dual Fermi gonna be made with such TDP's with the single version, I don't know.

don't expect 2x 512SP's in a dual part?

[neliz]

i just want to see at least 3 gf100 based geforce cards.

at $500, $600 and $700? two single parts at launch, cut down and (probably) full product at launch, although they could also launch a 256/320/384 or 448 sp part. Remember nvidia says "up to" 512 cuda cores.. and they don't want to be caught lying. dual fermi part is definitely a number of weeks after that.

I also wouldn't be surprised if the boards shown at CES were 256sp parts. Since most of the demos ran at 720p and even that simple rocket sled demo turned into a slideshow once they turned tessellation on, they must be hiding something.

Now that rajiff the thief (whatever) guy can either deny or confirm it, but in march we will see one hug *ss fermi based chip and one of more decent size. It would also be nice if the confirmed the three different part available in this and the next quarter. So far he seems to be telling me less than people that actually are under NDA.

[Blacky]

their single card (GTX380) won't gain their performance crown, so I wouldn't be surprised if they release a dual-gpu based fermi shortly after their fist single gpu cards, but lets see GTX 380 tdp and temps goes...

[Nedjo]

their single card (GTX380) won't gain their performance crown, so I wouldn't be surprised if they release a dual-gpu based fermi shortly after their fist single gpu cards, but lets see GTX 380 tdp and temps goes...

as annihilat0r pointed out there's just no space in 300W budget to squeeze two GTX380's... hell even two 360's would be challenge, and that would require significant mumbo-jumbo on binning side... but... they can always stitch something like they did with EVGA... you know GTX275+GTS250 freak... but this time with GTX360 and... err... GTS340!? They have know-how, and they'll have bunch of 360's if rumored yield is true

[annihilat0r]

Power draw is directly proportional to performance. What I don't get is that, if neliz is right and GTX 380 will draw 300W, then there can't be anything faster than that in the current node and architecture. Whatever they do to castrate Fermi to put 2 of it into 1 card, it must still be 300W. So I am guessing that Fermi won't be 300W but something like 240 or 250W. That way, a dual version might be at least somewhat faster than the GTX 380.

[Nedjo]

Power draw is directly proportional to performance. What I don't get is that, if neliz is right and GTX 380 will draw 300W, then there can't be anything faster than that in the current node and architecture. Whatever they do to castrate Fermi to put 2 of it into 1 card, it must still be 300W. So I am guessing that Fermi won't be 300W but something like 240 or 250W. That way, a dual version might be at least somewhat faster than the GTX 380.

or they can create card that would require completely new ATX power standard, and push it for broad acceptance

[neliz]

Power draw is directly proportional to performance. What I don't get is that, if neliz is right and GTX 380 will draw 300W, then there can't be anything faster than that in the current node and architecture.

A HD5970 also fits within the PCI-SIG specs, but can be easily pushed over.

[annihilat0r]

A HD5970 also fits within the PCI-SIG specs, but can be easily pushed over.

It can, manually. There won't be a version of a HD5970 which pulls more than 300W without any overclocking done on the user's part.