Do you have any experience working with CUDA on an enabled GPU that uses the PCI interface (as oppose开发者_如何转开发d to PCIe)? Does it work?
I cannot see a reason why it wouldn't work, and a quick google for "PCI CUDA" seems to confirm that it's possible. They do make some CUDA cards for PCI.
However, the bandwindth is one of the important bottlenecks on GPGPU, but probably not the most important one too (well, depends on the specific problem). If you're OK with having the Host<->Device transfer slower, then you're good to go.
精彩评论