I’m trying to do some profiling on unified memory with the nvprof profiler.
I am able to get the information about the page fault handling with -print-gpu-trace option but,
I saw the description that the Visual Profiler is able to tell me more detailed information such as migration reason or page fault access type.
I found this inside under link in “Profiler changes in CUDA 9.0 section”.
Is there any possible way to find such information without using Visual Profiler?
Since Visual Profiler gets the profiling results of nvprof as an input, I assume there must be existing detailed information about of page faults inside of the nvprof profiling results.
But I don’t know how can I see it without the Visual Profiler…
If anyone knows about this pls help me out.