University of Antwerp Makes 4k € Supercomputer with Four GeForce 9800 GX2 Cards

[Cold Storm]

Hey it takes a lot to power that beast! lol

 

[lemonadesoda]

If you are interested in this sort of stuff, head over to http://www.gpgpu.org/

A review of math accelerators would be interesting, esp. performance comparison (raw FPU) for

1./ Physx
2./ CUDA
3./ Clearstream
4./ SpursEngine
5./ ...

 

[jbunch07]

Hey it takes a lot to power that beast! lol

i wonder if they could have got away with 1200w or even 1000w?

 

[Cold Storm]

I don't think so. Corsair's little calculator says I need 1000w just for 2 9800gx2...

and the PSU Calculator only goes up with two in sli... so who knows?!?!

 

[jbunch07]

I don't think so. Corsair's little calculator says I need 1000w just for 2 9800gx2...

and the PSU Calculator only goes up with two in sli... so who knows?!?!

hmm well the only other psu that is higher than that that i know of is the ultra 1600w

but hey...if you got to have it... why not.

 

[muon]

If each card uses say 200w, 4 of them means 800w. Add all the other components and also overclock, a 1500w psu is probably advisable.

 

[jbunch07]

If each card uses say 200w, 4 of them means 800w. Add all the other components and also overclock, a 1500w psu is probably advisable.

true true.
i forget we are talking about 8gpus and 4cpus.

 

[Bull Dog]

And of course what is even funnier is that its running on MSI's 790FX board (only board I know of that would support such a setup).

 

[Cold Storm]

yeah, there we go! sounds good to me also.

 

[jbunch07]

And of course what is even funnier is that its running on MSI's 790FX board (only board I know of that would support such a setup).

haha true that.
this is good advertising for MSI

 

[wolf]

the great thing about the 8 gpu setup means that a healthy shader overclock ober 8 cores means decent gains in overall computational power.

stock shader clock 1500mhz X 128 SP X (MADD (2 FLOPs) + MUL (1 FLOP) X 2 (because theres 2 Gpu's) = 768

GeForce 9800 GX2---------------768---G/Flops---[1152 (peak theoretical)]
GeForce 9800 GX2 SLI (x4)------1536--G/Flops---[2304 (peak theoretical)]
GeForce 9800 GX2 SLI (x8)------3072--G/Flops---[4608 (peak theoretical)]


An over clock of the shader units to somewhere in the order of ~1750mhz, which i would argue is modest on a G92, would bring you these figures.

GeForce 9800 GX2------------896---G/Flops---[1344 (peak theoretical)]
GeForce 9800 GX2 SLI (x4)----1792--G/Flops---[2688 (peak theoretical)]
GeForce 9800 GX2 SLI (x8)----3584--G/Flops---[5376 (peak theoretical)]

so we can see that a modest overclock here brings you a half a teraflop jsut like that, not bad. and a shader overclock of 2000mhz would give you this nice figure on this sytem.

GeForce 9800 GX2 SLI (x8)----4096--G/Flops---[6144 (peak theoretical)

also interestingly ive made some figures based on the current GTX280 estimates.

1296mhz x 240 SP X (MADD (2 FLOPs) + MUL (1 FLOP) X 4 (because we would have four in this system) = 2488 G/flops [3732 (peak theoretical)]

now overclocking....

~1500 mhz x 240 SP X (MADD (2 FLOPs) + MUL (1 FLOP) X 4 (because we would have four in this system) = 2488 G/flops [4320 (peak theoretical)]
~1750 mhz x 240 SP X (MADD (2 FLOPs) + MUL (1 FLOP) X 4 (because we would have four in this system) = 3360 G/flops [5040 (peak theoretical)]
~2000 mhz x 240 SP X (MADD (2 FLOPs) + MUL (1 FLOP) X 4 (because we would have four in this system) = 3840 G/flops [5760 (peak theoretical)]

just imagine four GX2-280's... a man can dream......



The Tech Report

One thing I should note: I've changed the FLOPS numbers for the GeForce cards compared to what I used in past reviews. I decided to use a more conservative method of counting FLOPS per clock, and doing so reduces theoretical GeForce FLOPS numbers by a third. I think that's a more accurate way of counting for the typical case.

Wikipedia

For example the GeForce 8800 GTX has 518.43 GigaFLOPs theoretical performance given the fact that there are 128 stream processors at 1.35 GHz with each SP being able to run 1 Multiply-Add and 1 Multiply instruction per clock [(MADD (2 FLOPs) + MUL (1 FLOP))×1350MHz×128 SPs ＝ 518.4 GigaFLOPs][4]. This figure may not be correct because the Multiply operation is not always available[5] giving a possibly more accurate performance figure of (2×1350×128) = 345.6 GigaFLOPs.

 

[DrPepper]

All that money and they have to sit the screens on 2 packs of A4.

 

[jbunch07]

All that money and they have to sit the screens on 2 packs of A4.

haha i saw that to and tried to ignore it.

 

[DrPepper]

Would be cool if they could use CUDA to run crysis or 3dmark Vantage

 

[DrPepper]

http://fastra.ua.ac.be/images/pic_tones_large.jpg I thought it was a phenom or is that more crafty intel advertising.

 

[jbunch07]

ha!
Intel is just wishing!

 

[DrPepper]

That would really benefit if it could use pci-e 3 it says on the website that the pci-e was a bottleneck.

 

[Morgoth]

so this doest need a a sli chip?

 

[DrPepper]

so this doest need a a sli chip?

no

 

[Morgoth]

just readed there website
and shecked there specs when nehalem server parts are comming out it will easly crush that system

 

[Fleekar]

Hurray for using xp haha

 

[DrPepper]

just readed there website
and shecked there specs when nehalem server parts are comming out it will easly crush that system

I doubt it since gpu's are much more efficient for that kinda stuff. Unless you mean they replace the phenom with a nehalem.

 

[eidairaman1]

You could turn water into steam with the GPU temps. Most definitely not worth the cash for the Enthusiast Gamer.

Have you ever thought that gaming parts like NVIDIA's GeForce 9800 GTX video cards for instance, can be used for building a supercomputer. Maybe no, but researchers at the University of Antwerp in Belgium have proven that it's possible to build one. Using four NVIDIA GeForce 9800 GX2 graphics cards, AMD Phenom 9850 processor, 4x2GB Corsair Twinx DDR2 PC6400 memory and MSI K9A2 Platinum motherboard, FASTRA costs less than 4000EUR to build and thanks to NVIDIA's CUDA technology and delivers roughly the same performance as a supercomputer cluster consisting of hundreds of PCs. This new system is used by the ASTRA research group, part of the Vision Lab of the University of Antwerp, to develop new computational methods for tomography. Tomography is a technique used in medical scanners to create three-dimensional images of the internal organs of patients, based on a large number of X-ray photos that are acquired over a range of angles. ASTRA develops new reconstruction techniques that lead to better reconstruction quality than classical methods. You can read more about the FASTRA GPU SuperPC project over here.



Source: DVHardware

 

[DrPepper]

You could turn water into steam with the GPU temps. Most definitely not worth the cash for the Enthusiast Gamer.

Then use the steam to turn turbines to supply enough power to light a small village

 

[eidairaman1]

Then use the steam to turn turbines to supply enough power to light a small village

LMAO!!! But the energy it takes to run those cards vs the energy produced is not that great