In CUDA SDK i found example which uses OpenGL and CUDA to modify VBO on GPU. When we use single graphic card it’s simple but I want use two. One GPU + CUDA to create/modify VBO (procedural terrain), next step - copy it to second card and draw scene (using one GPU of course). Is possible to copy this VBO from first card to second - only from card to card (not from card to CPU and from CPU to second card)?
Looks like here’s no answer but I have another question.
I have two OpenMP sections - first to draw, second is waiting for use CUDA. When I only draw the scene, I have about 200-300fps. Second section is waiting for pushing button - it starts CUDA copmuting. And here’s the problem - sometimes it works (using CUDA) fast and I have same number of fps, sometimes slow - it works longer and i have about 120fps so probably one GPU is using for both operations - drawing and computing. To choose GPU for CUDA, I use
cudaSetDevice(1);//(0)
I changed 1 for 0 - and nothing - sometiems is good, sometimes not. I switched off SLI mode but it doesn’t help. Does anybody know how to select one GPU for GL and another for CUDA correctly?