I have two x16 PCI Express slots. I use now NIVIA Quadro FX1500 graphics card.
Can I upgrade to graphic cards suitable for CUDA? Or is this type of slots fully outdated?
Please give info because I have no idea how to implement CUDA, with which graphics card?
Err… PCIe x16 is the best GPU interface there is. The only issue is gen1 vs gen2. For a list of supported GPUs, look at NVIDIA’s list. Personally, I would stick with GPUs listed as Compute Capability 2.0 or higher, since they have far fewer sharp edges.
16 lanes connected to a gen1 slot nets you 4GBps of bandwidth, while 16 lanes connected to a gen2 slot nets you 8 GBps of bandwidth.
Even if you do have gen1 x16 PCIe slots on your MOBO, 4GBps of bandwidth is plenty for a GTX 470. Only dual-GPU cards like the GTX 295 or the Radeon 5970 are slightly bottlenecked in a x8 gen2 slot (or a x16 gen1 slot), as the bandwidth the single slot provides is split between the GPUs. What that means is that if you install a GTX 295 into a x16 gen1 slot, each GPU will only have access to 2GBps of bandwidth, which could hinder them a bit depending on the app.