I would like to know what is computer configurations with GTX 295 to implement massively parrallel computing with CUDA? Memory? CPU? Motherboard? Power source?
It is in the case of Graduate School project. So, we have to borrow the computer which is compatible with GTX 295 and High performance with CUDA.
Thank you
Noboby can tell me what are system needs to correctly use GTX 295 with Cuda? I know that I have to have a motherbord with PCI Express X16 but I don’t know if I have to have a system configuration with a specially speed of CPU, of RAM, etc.
Thank you for your answer !
You don’t need hardly any special configuration. AFAIK, you need a motherboard with PCIe x16 and at least 680 W power supply. I would recommend Intel i7 architecture since it is a recent technology but if you have lack of money any brand-new motherboard with CPU will do the job. I would also recommend a decent PC case since hi-performance GeForce cards need a good air circulation to be cooled. (And you need two slots on the back side of a case). Concerning RAM, well maybe even 512 MB would work but why would you have less CPU than GPU memory? :)
If you want to make use of the two GPUs in a GTX 295 card, you want at least a dual-core processor. Also, for best transfer speed between the CPU and the card, you want PCI Express 2.0. The GTX 295 can draw up to 289W, which means that it requires both a 6-pin and an 8-pin PCI-Express power connection. Many power supplies lack the 8-pin connector, so typically a 2x6 to 8-pin adapter is included with the GTX 295. So you will want a power supply that is at least 650W with 6-pin + 8-pin connectors or three 6-pin connectors.
Also, as was pointed out, the card is wide enough to occupy two slots, so make sure there is enough room in the case for that.
Thank you for your answer !!!
My last question is about the the architecture of the OS. What will be the different between a 32bit and a 64 architecture to compute massively parrallely on the GPU thanks to CUDA ?
Both 32 and 64 bit should be fine for CUDA. On the CPU side, 64-bit x86 code generally runs a little faster because the instruction set has more available registers than 32-bit x86.
Just to give a short warning: So far CUDA works only with a single GTX 295 under Vista and Win7.
Those OSes will only detect all GPUs of the first display card, but just 1 GPU per card for each additional card.