Hi, i’m new with CUDA and going to get a GTX 295 :D . This card does have two (SLI connected) GPU’s. Does this card use both of the GPU’s to calculate the CUDA FFT or other algorithms ?
It’s not going to happen automatically, but you can send an FFT to each GPU and they will run in parallel.
Ok thats fine. Is there any code sample? or can you tell me how to send a command to a certain GPU?
Check out the following SDK examples: simpleMultiGPU, cudaOpenMP, MonteCarloMultiGPU
These samples can not answer this question.
Unlike CUBLAS, there is no such function that likes cublasInit() in CUFFT. Is anyone can figure out how the CUFFT can execute on multiple GPU seperately?
Do you mean if we use cuCtxCreate() function with different device number in different threads or processes, the FFT functions will automatically be sent to the relative GPU device?