Intervals of GPU activity will be shown for any graphics, just select "Analyze DirectX pipeline events" option in any algorithm analysis type. When result opened you should be able to see GPU activity on "Graphics" tab. Thread and Process based groupings in grid also will have "GPU Engines Usage" metric.
Option "Analyze Processor Graphics hardware events" is for Intel HD Graphics only. Predefined GPU metrics groups allow to monitor execution units activity and either analyze sampler, general memory and cache accesses or track accesses to different types of GPU memory.
Both options are available on Windows only at this moment.
Just one more quick question. If I have a function that calls a "draw" function, for example, and the draw function utilizes the GPU and the caller function is called from CPU, will the data collected show "idle" for CPU time (ie the CPU is idel while the GPU works) so that we can deduce the GPU time (or at least that the CPU is waiting on some other resource) or will this time not be accounted for at all?
For GPU profiling/monitoring you can use Cuda toolkit.It is supported on Linux.I am using it to test Cuda programs. I am on Windows.
For OpenGL profiling there is perfKit but it seems to work only on Windows.
I downloaded and installed the cuda toolkit but it's too poor and the documentation is very bad. I was unable to follow the Visual Profiler instructions and get NSight to profile my application.
Joe, depending on actual CPU behavior after submitting work to GPU VTune will show activity, wait or spin within the GPU interval right on submitting CPU thread. Again, at this moment GPU activity may be shown for WIndows only.
Through we may add Linux support for Intel Integrated Graphics in the future.
I do not have any problems with cuda toolkit,but I am writing mainly console based programs. Did try to run and profile a few important benchmarks test tools like deviceQuery.?Did you check the full compatibility of your GPU? Try to update your drivers to the newest one.