The GT300/Fermi Thread

[ajaidev]

Data comes in over InfiniBand, works its way into main memory and is copied; before it is moved down to the GPU, it is copied again and that copy is what is moved. The driver changes allow for the data moved into memory to be moved down to the GPU in one fell swoop.What do we get from this ? They have been able to boost a 30% increase in speed.TheTesla 20 GPU co-processors & the appliances based on them will be available in the 2nd quarter of 2010. The GeForce graphics cards based on the same GPU chips will be seen around first quarter.


InfiniBand is not new news...

[Luka_Aveiro]

Universidade de Aveiro teaching Cuda.

Serious business.

[Sh1tyMcGee]

any word on size?

shhhhhhhhhhh dont tell Nedjo, he works for AMD haha j/k

[LedHed]

Universidade de Aveiro teaching Cuda.

Serious business.

but seriously every top Engineering college in the US is on the list.

[Solus Corvus]

Yeah, get all those youngsters hooked on cuda. That will help standards adoption...

[zerazax]

but seriously every top Engineering college in the US is on the list.

What's their criteria though? All those universities teach multiple classes - if it's just one class, or one research group teaching it, does it mean it's 'widely taught'?

If that were the case, then MATLAB would be the world's most taught program in higher institutes

Anyways, I'm a bit dissappointed they didn't go over the GeForce's too much. I know GPGPU is their new thing but I'd hate for them to put gamer's as an afterthought

[JohnZS]

3gb Fermi Quadro shipping in Early Q2 2010.Double Precision Performance = 520GFlops 190W TDP

6gb Fermi Quadro shipping in Q3 2010. 630 GFlops Double Precision Performance TBD TDP

Dimensions 9.75" x 4.376", Dual Slot

Supports CUDA, C/C++/Fortran, OpenCL, DirectCompute, Physx, OpenGL 3.2, DirectX 11

[N19h7m4r3]

3gb Fermi Quadro shipping in Early Q2 2010.Double Precision Performance = 520GFlops 190W TDP

6gb Fermi Quadro shipping in Q3 2010. 630 GFlops Double Precision Performance TBD TDP

Dimensions 9.75" x 4.376", Dual Slot

Supports CUDA, C/C++/Fortran, OpenCL, DirectCompute, Physx, OpenGL 3.2, DirectX 11

Is this accurate? If so 9.75" is not bad at all for a such powerfull card.

One would hope then that the geForce equivalents are slightly shorter.

[003]

We will definitely be seeing the GeForce over 200W TDP as it will be clocked higher than both Quadro and Tesla, although ECC will be disabled.

[trinibwoy]

190W TDP @ such low clocks doesn't bode well though. I wonder if they're setting Tesla expectations based on A2. Given the non-linear increase in power consumption with clock speed (and voltage) there isn't much room to grow.

[Cybercat]

190W TDP @ such low clocks doesn't bode well though. I wonder if they're setting Tesla expectations based on A2. Given the non-linear increase in power consumption with clock speed (and voltage) there isn't much room to grow.

They haven't gotten A3 back yet have they? So yes I'd expect that everything is based on A2 right now, and that might be a contributing factor to the Tesla clocks being so low. My guess is they were under pressure to finalize specs for their buyers ahead of all other product lines, so A2 was what they were forced to go on.

You have to also consider that these cards are using exponentially more memory than consumer cards, raising overall TDP.

[trinibwoy]

That's a fair assumption. Would be very interesting to know if they're actually running A2 wafers now though.

[highoctane]

That's a fair assumption. Would be very interesting to know if they're actually running A2 wafers now though.

Man I would sure assume they are if only from a development & testing standpoint until A3 is ready. Wasn't there some chatter about them getting the development kits or the likes out soon?

[003]

It sounds like you lack some basic logical reasoning abilities. There is plenty of information about how these firms design chips, and the resulting power specs...maybe you are in denial because nVidia has been selling in a box for the last year? I suggest you go read the tech specs of RV800 (can you find those on your own?) and then read about the JumboDie TM design scheme that nVidia thinks is going to work (and is producing record low yields). A basic understanding of chip physics will tell you that 5890 is definitely going to beat whatever nVidia comes out with in terms of performance/watt. With such crappy yields and ATIs headstart (and opportunity to die shrink and cut prices) they are going to be also losing in performance/dollar. At this point the only thing that nVidia could possibly win at is total XOPS, which considering the differing technologies may be better for WCG etc. at performance/card, but that is a relatively useless metric in any market that matters...

Once again you have no idea what the performance of the high end Fermi based GeForce will be. I actually know some information that is under NDA and it will be considerably faster than the RV870, I can promise you that. Based on the information about Quadro and Tesla you can make a ballpark guess of what the TDP will be but nothing specific. Additionally, current TDPs are likely based off of A2 silicon which will not be the production silicon, further complicating the ability to make any kind of accurate guess.

I'm sorry but with zero information about performance and only very rough guesses about TDP, there is no logical way you can know anything meaningful about performance per watt. Trying to claim otherwise makes you come off as a fool.

Furthermore, so what if perf/watt happens to be lower than that of the 5870? It will definitely be higher than the 5970, that's for sure. Anybody seriously considering buying a gaming card like this isn't going to care if TDP is 190W or 230W (230W was pulled from my arse). They will care about price and performance, both of which Fermi is going to deliver in a big way (and no trolls, I do not mean price will be big).

[Solus Corvus]

You mean like 40% faster? Even if it is 40% faster then a 5870 that won't put it ahead of a 5970. And if as you say only price and performance matter then AMD only has to reduce the price of the 5970 to be competitive.

I want to see what the dual fermi boards do, but I suspect we won't be seeing those for a while after the single gpu boards launch.

[nullface]

I care, or at least i care (A LOT) about my idle power consumption, and if Fermi pulls twice as much power idle than HD5800 its a no brainier for me, regardless of performance difference.

And im a bit confused about the ECC. One moment you use it as a feature to hype both Quadro and GeForce and the next you say it will make no difference if GeForce has it or not (which it wont).
Im getting the impression that nvidia is pulling out a lot of technology that looks good on paper, but might not be that usefull IRL, or at least in gaming, which must be the biggest high end GFX market.

[003]

I care, or at least i care (A LOT) about my idle power consumption, and if Fermi pulls twice as much power idle than HD5800 its a no brainier for me, regardless of performance difference.

And im a bit confused about the ECC. One moment you use it as a feature to hype both Quadro and GeForce and the next you say it will make no difference if GeForce has it or not (which it wont).
Im getting the impression that nvidia is pulling out a lot of technology that looks good on paper, but might not be that usefull IRL, or at least in gaming, which must be the biggest high end GFX market.

There is literally no difference between idling at 10W and 20W and that would be a seriously stupid reason to avoid a considerably faster GPU. Nvidia has always been very good with idle power though. IIRC the GTX280 idles at less than a 4870. I've actually seen reviews that have opposite claims. Probably depends on how they are tested.

[Solus Corvus]

What happened last generation doesn't necessarily apply to this generation. The 5xxx series has vastly improved power consumption, and not just because of the new process. How will Fermi compare to that? We won't know until launch unless someone with information under NDA decides to tell us.

[zerazax]

We won't know until launch unless someone with information under NDA decides to tell us.

Well with all the folks on this board claiming to know information under NDA that shouldn't be hard

[mindfury]

3gb Fermi Quadro shipping in Early Q2 2010.Double Precision Performance = 520GFlops 190W TDP

6gb Fermi Quadro shipping in Q3 2010. 630 GFlops Double Precision Performance TBD TDP

Dimensions 9.75" x 4.376", Dual Slot

Supports CUDA, C/C++/Fortran, OpenCL, DirectCompute, Physx, OpenGL 3.2, DirectX 11

190W is typical power consumption,not TDP.

Max power consumption=225W

[003]

Well with all the folks on this board claiming to know information under NDA that shouldn't be hard

Unless of course they don't want to get the people who breached NDA in trouble.

[skugpezz]

Still waiting to see if fermi is real....

[Solus Corvus]

Unless of course they don't want to get the people who breached NDA in trouble.

Not telling and not knowing have little difference to the rest of us. Either way it doesn't further our discussion of the product.

[JohnZS]

Still waiting to see if fermi is real....

Oh the Fermi is real! Just it's going to be a long wait if the Quadro Cards are not coming until Q2 2010

003 I am not sure if the Quadro Fermi clocks will be lower? (granted they might be on the memory), I am going by the Quadro FX5800 which had a Core Clock of 650Mhz compared to the GTX 280 which had a core clock of 600Mhz

Now, BFG will most likely do an OCX, Gainward a Golden Sample and EVGA a WTF, BBQ FTW edition with higher clocks, but the fact remains we have no idea whether the Quadro cards will be clocked higher, lower, or the same as the consumer cards.

John

[Teemax]

I actually know some information that is under NDA and it will be considerably faster than the RV870, I can promise you that.

First you claimed you cannot "break" NDA.
Immediately after, you "broke" NDA by claiming that Fermi is significantly faster.

To me, this sounds like a lot of hot air. If you want to be taken seriously, back up your claims with at least a tiny bit of evidence.

Or just drop the silly disguise that you know something more than then rest of us mortals, mr. Fanboy.