Error: Internal profiling error 4075:999 when using --analysis-metrics


I am trying to profile a kernel. Apparently it works well, and gives the expected output. I am unable to find any problem using both cuda-gdb and cuda-memcheck. I can profile it using this commands:

nvprof  ./test
nvprof -o test2.nvvp ./test

but when I try this:

nvprof --analysis-metrics -o test.nvvp ./test

It stops with the error:

==23785== Some kernel(s) will be replayed on device 0 in order to collect all events/metrics.
==23785== Warning: PC Sampling is not supported on the underlying platform.
Replaying kernel "void kernelDecompress<float>(float*, unsigned int*, unsigned char const *, int const *, long const *, int, int, int, int)" (3 of 52)... 
	4 internal events
==23785== Error: Internal profiling error 4075:999.

If I try with the options:

nvprof --metrics all -o test2.nvvp ./test

It ends with the same error.

I have been unable to find this error in the forums, is there a known workaround to this?

I am using a Tesla k80 card, and cuda version 10.1.105.

Thanks for your help.

Hello, did you find a solution to your problem? I have exactly the same problem as you.


I did not find a solution. I started using NSight and newer GPUs I was waiting for.

Sorry for not being more helpful.