Reputation: 867
I'm trying to profile my application that uses cuBLAS exclusively with Nvidia Visual Profiler on Windows, however it shows that there's no GPU usage in my application at all! That is, the timeline is completely empty except for profiling overhead. Just to make sure someone didn't change out the security settings or something under me, I profiled an application with a kernel and cudaMemcpy
calls, and it is profiled properly. What gives? Am I missing a setting? Linking to the wrong version of cuBLAS libraries? Or is it not actually calling the GPU (although seems completely improbable to me...)? I'm using Intel's compiler for 64-bit support, if that makes a difference.
Thanks!
Upvotes: 4
Views: 482
Reputation: 867
For anyone who runs in to this problem in the future: I had to use cudaProfilerStart()
and cudaProfilerStop()
around my cuBLAS function. Just adding cudaDeviceSyncrhonize()
didn't fix the problem.
Upvotes: 5