PCIe Impact PCIe 1.0 vs. 2.0 performance impact.

I’m getting started with CUDA by using my existing L1N64-SLI/WS motherboard as a cheap starting point. I’m thinking of installing four BFG GTX 280 H2OC cards into the PCIe 1.0 slots (2@8x, 2@16x).

My question to the knowledgeable is will I ever come up against a PCIe 1.0/2.0 saturation issue (assuming my code ever gets that good)? My intent is to go RHEL 5.1 (non-graphical, console-mode only).

Thanks in advance,


Sure, if your code is

  • copy 900 megs (or so much that you can’t pipeline memory copies)

  • run a very short kernel

  • copy 900 megs back

repeated a lot. In that case, PCIe 2 will net you a huge perf boost. Is that a likely scenario? I doubt it.

Excellent, this means I should be able to get quite alot of mileage out of this set-up.