question about multiple GPUs

[shingoxlf]

in the CUDA_SDK, there is a example called “simpleMultiGPU”.

Here is how it uses multiGPUs

for(i = 0; i < GPU_N; i++)
threadID[i] = cutStartThread((CUT_THREADROUTINE)solverThread, (void *)(plan + i));
cutWaitForThreads(threadID, GPU_N);

My question is these threads in CPU are parallel or not?

It seems like they just sequentially call the function “solverThread” in which the kernel function is involved.

Anyone can help?

[eyalhir74]

in the CUDA_SDK, there is a example called “simpleMultiGPU”.

Here is how it uses multiGPUs

for(i = 0; i < GPU_N; i++)

threadID[i] = cutStartThread((CUT_THREADROUTINE)solverThread, (void *)(plan + i));

cutWaitForThreads(threadID, GPU_N);

My question is these threads in CPU are parallel or not?

It seems like they just sequentially call the function “solverThread” in which the kernel function is involved.

Anyone can help?

Look at the cutStartThread method implementation - it is a wrapper around the WinAPI CreateThread function

call, which indeed starts a new CPU thread. The solverThread method is the function called by each CPU thread.

eyal