Supercomputer based on four video card NVIDIA GeForce 9800 GX2

Researchers working at the University of Antwerp created a supercomputer based on four video card NVIDIA GeForce 9800 GX2. The price of the system was less than 4,000 euros, but the performance is comparable with a cluster of hundreds of PCs, writes iXBT.

The supercomputer uses a research group ASTRA, whose task is to develop computational methods for tomography. According to the band members, eight of NVIDIA GPUs provided Bole performance than 300 Intel Core 2 Duo, clocked at 2, 4 GHz. The new system solves problems faster and potreblinet energy is much less than the supercomputer cluster.

To build a system, which received the designation FASTRA, it was chosen as the platform AMD, because scientists have been unable to find a motherboard Intel, able to take four the GeForce 9800 GX2. Interestingly, SLI technology is used, because the interaction between the GPU is not required. Instead, applications use programming model NVIDIA CUDA, which enables parallel operation of all eight GPU.

Technical data of the system:

The motherboard - MSI K9A2 Platinum;
Processor - AMD Phenom 9850; Cooler -Scythe Infinity;
The memory - 4x2 GB Corsair Twinx DDR2 PC6400;
Video card - 4x MSI 9800GX2;
Winchester - Samsung Spinpoint F1, 750 GB;
Power supply - ThermalTake Toughpower, 1500 W;
Housing - Lian-Li PC-P80 Armorsuit.
Powered supercomputer running a 64-bit version of Windows XP.

The biggest challenge faced by the creators of the system, it was cooled to four graphics cards. To lower the temperature of the GPU, had to remove the side cover. The idle temperature is 55 ° C, under full load and increases to 86 ° C, while in the case of increasing the frequency of shader 20% above the nominal value - up to 100 ° C.

Rate possibility FASTRA allows lower illustration showing the time spent on the reconstruction of one image data received from the scanner. The upper value corresponds to the GPU on a higher frequency, the average - work in a normal mode, and the bottom - an indicator of supercomputer CalcUA, consisting of 256 nodes on the AMD Opteron processor 250 (2, 4 GHz), the purchase of which in March 2005, has cost the university 3, 5 million. euro.


See also

New and interesting


We on Google+