Cryptohaze.com

Hello,

Currently I do not have a CUDA supported GPU and have been looking to acquire a new hardware setup that is supported so that I may further fuel my interest for this technology. I have the following questions regarding Nvidia GPU's and a systems CPU.

I do not have the desire to purchase a full sized desktop machine for these purposes and have been looking at the mini-ITX offerings from Zotac. They currently have available the following two models of mini-ITX boards, one with a dedicated CPU and the other with a CPU socket and one PCI Express x16 expansion slot.

IONITX-A-U
Atom 330 1.6ghz dual core CPU (soldered)
GeForce 9400M
MCP7A-ION Chipset

http://www.zotacusa.com/zotac-ionitx-a-u-atom-n330-1-6ghz-dual-core-mini-itx-intel-motherboard.html

GF9300-D-E
Intel socket LGA775 (Intel Core 2 Duo, Core 2 Quad, Pentium and Celeron up to 1333 MHz front-side bus)
GeForce 9300
1 x PCI Express x16

http://www.zotacusa.com/zotac-geforce-gf9300-d-e-itx-wifi-lga-775-mini-itx-intel-motherboard.html

My questions being, how important is a systems CPU performance in regards or relation to the GPU? Will a slower CPU bottleneck the performance of the GPU? Are the GeForce 9400M and GeForce 9300's a viable solution for such applications as the cuda-multiforcer?

Any insight would be greatly appreciated.

Thank you.

The 9400M will run the CUDA Multiforcer - not fast, but it will run it. I believe it's slower than a CPU, though.

CPU vs GPU performance is entirely dependent on the app. Some apps use the CPU and GPU separately, some (the CUDA multiforcer currently) use the GPU almost entirely with the CPU only handling feeding (and busy waiting in the older CUDA SDK versions), and some heavily integrate the CPU/GPU processing. It just depends on the app.

A 9400M is not a powerful GPU, though, so you won't have any trouble with any reasonable CPU. Personally, I'd put a GTX260/216SP in the PCI-E slot if you're serious about doing CUDA work.