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

Saturday, May 31 2008

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

posted by malware - 10:36 AM | Related News |

User comments

by tkpenalty (May 31st - 10:58 AM) - Reply

Hmm... Octa G92s... thats almost two Teraflops isnt it? Supercomputer in a full tower microcomputer. Kewl. Then again GPUs have WAY more number crunching power (FP calc speed) over normal microprocessors...

Thats pretty cheap for a supercomputer.

by Conti027 (May 31st - 11:08 AM) - Reply

looks like the cards would over heat

by Weer (May 31st - 11:10 AM) - Reply

I didn't know Octa-SLi could work.

by tkpenalty (May 31st - 11:14 AM) - Reply

by: Weer
I didn't know Octa-SLi could work.

With CUDA it will allow it. I Don't think that they are running in SLi, they are just plugged in and working as processors.

by Basard (May 31st - 11:17 AM) - Reply

they should set it up to balance the budget.

by Gam'ster (May 31st - 11:22 AM) - Reply

Thats some nice gear someone there has an eye for good tech lian-li case, nvidia gpus, but i expected an intel proc but rock on AMD :rockout:
And its being used for medical research ( a damed good thing ) not some useless crap well never see or hear of again.

Gam

by Weer (May 31st - 11:28 AM) - Reply

by: tkpenalty
With CUDA it will allow it. I Don't think that they are running in SLi, they are just plugged in and working as processors.

How could it work in full power in CUDA if it's not running in SLi? It will only use the card connected to the monitor(s).

by Mussels (May 31st - 11:58 AM) - Reply

by: Weer
How could it work in full power in CUDA if it's not running in SLi? It will only use the card connected to the monitor(s).

faulty logic there. SLI allows two images onto one display, but the work they're doing isnt even 3D, nor does it need to be displayed. They're using them as number crunchers through CUDA, not as video cards.

by tkpenalty (May 31st - 12:28 PM) - Reply

by: Mussels
faulty logic there. SLI allows two images onto one display, but the work they're doing isnt even 3D, nor does it need to be displayed. They're using them as number crunchers through CUDA, not as video cards.

I thought he would have the sense... u finished off what I was saying.

Yes. With CUDA... 9800GX2 become the equivelent of stacks of CPUs. The thing that is awesome about this is the fact that 4000 euro is MUCH cheaper than a supercomputer equivalent -COUGH COUGH NASA'S SUPERCOMPUTER COUGH-

by DanTheBanjoman (May 31st - 1:04 PM) - Reply

by: Weer
I didn't know Octa-SLi could work.

Multithreading using multiple GPU's doesn't require SLI. At work we utilize multiple GPU's without the need of SLI/Crossfire as well. In fact, we can't use either of those.
In theory it could even run on a ATI+NV card.

by Ripper3 (May 31st - 1:04 PM) - Reply

by: Conti027
looks like the cards would over heat

The GX2 sucks in air from top bottom and sides, and exhaust it out the back, so it should be alright.

I wonder how well this machine would run if it was running on either dual Opterons or dual Xeons. But if it's able to do what would usually take hours in seconds, I'm sure they don't need any more power that it has now.

Actually, just goes to show the raw, unused power some people have in their machines. Of course, I don't think anyone on these forums has four GX2s in one machine, but the 2900s and 3800s do well at F@H already.

by Jizzler (May 31st - 2:14 PM) - Reply

Naw, just build another one. :)

Could use multi-socket boards, PCIe expansion systems, etc, but the cost would start to climb considerably compared to a cluster of these boxes.

by Ripper3 (May 31st - 2:23 PM) - Reply

Just remembered that Nvidia has this: http://www.nvidia.com/page/quadroplex.html
I'm sure it's a more expensive solution, and probably not as versatile as four GX2s.

Yeah, jsut checked, definitely cheaper to buy four GX2s and a system to put them in.
One Quadro Plex Model IV costs $10750 :O and it's in backorder.
States it can be used for more than jsut rendering, but for the price, and the potential waiting time, I'd just build a system with four GX2s.
Oh, and it only has two GPUs.

by Random Murderer (May 31st - 2:35 PM) - Reply

by: tkpenalty
With CUDA it will allow it. I Don't think that they are running in SLi, they are just plugged in and working as processors.

that's how i see it as well, instead of actually using the cards for graphics rendering, they're using each g92 core for floating point calculations, just like f@h does with ati cards(and nvidia cards soon)

by Grimskull (May 31st - 2:40 PM) - Reply

wonder what it would be like to play solitaire on??? bet its just only meh!! :laugh:

by Cold Storm (May 31st - 2:40 PM) - Reply

by: Random Murderer
that's how i see it as well, instead of actually using the cards for graphics rendering, they're using each g92 core for floating point calculations, just like f@h does with ati cards(and nvidia cards soon)

Thats the same thing I get out of it as well... and Can't wait for Nvidia cards to work with F@H!!!

by Mussels (May 31st - 2:42 PM) - Reply

by: Cold Storm
Thats the same thing I get out of it as well... and Can't wait for Nvidia cards to work with F@H!!!

i agree. G92 cores have a lot of processing power, whilst running cool (on anything NOT the reference cooler :P )

by Cold Storm (May 31st - 2:45 PM) - Reply

by: Mussels
i agree. G92 cores have a lot of processing power, whilst running cool (on anything NOT the reference cooler :P )

That is true. only thing is that my gx2 is only cooled by that right now! till I find a cooler for it!

by WarEagleAU (May 31st - 3:12 PM) - Reply

Awesome for the Phenom proc.

by tkpenalty (May 31st - 3:17 PM) - Reply

by: Mussels
i agree. G92 cores have a lot of processing power, whilst running cool (on anything NOT the reference cooler :P )

Actually the 9800GX2 temps are cooler than Reference 8800GT temps, by 10*C most of the time.

by v-zero (May 31st - 3:17 PM) - Reply

I wonder if this wouldn't work better with four 3870X2s. Theoretically the 3870s are just as powerful, and are cheaper to boot (and made on a better manufacturing process). Considering that the weakness of the 3870 architecture in gaming is mostly its AA implementation I think it may be a more powerful, and certainly more parallel raw-computing design.

Very interesting though.

by Mussels (May 31st - 3:34 PM) - Reply

by: v-zero
I wonder if this wouldn't work better with four 3870X2s. Theoretically the 3870s are just as powerful, and are cheaper to boot (and made on a better manufacturing process). Considering that the weakness of the 3870 architecture in gaming is mostly its AA implementation I think it may be a more powerful, and certainly more parallel raw-computing design.

Very interesting though.

well since they ran their program on CUDA, theres no information on whether their needs could even be met on ATI.

he main reason why they configured an AMD system is because they couldn't find a motherboard for the Intel platform that could fit four GeForce 9800 GX2 graphics cards. Another interesting note is that this system doesn't need SLI, their application uses the NVIDIA CUDA programming model which makes all eight GPUs work in parallel. The researchers say they don't need SLI during a reconstruction as every graphics card communicates directly with the CPU, no inter-GPU communication is needed.

by v-zero (May 31st - 4:19 PM) - Reply

by: Mussels
well since they ran their program on CUDA, theres no information on whether their needs could even be met on ATI.

AMD have a "software stack" for stream computing also...

by jbunch07 (May 31st - 4:40 PM) - Reply

haha i like the 1500w psu.
good c/m as well.
im betting that the cards in the middle get nice and warm though.

anyway this things is very B/A

by Cold Storm (May 31st - 4:45 PM) - Reply

Hey it takes a lot to power that beast! lol

Post your own comment

Username:
Password:
Register Here!