Hello,
When I try to profile an application that uses 30 cuda streams using nvprof, it gave me an error like this "Error: Internal profiling error 3072:999". I am able to profile with 18 stream though.
Thanks,
Ray
Hello,
When I try to profile an application that uses 30 cuda streams using nvprof, it gave me an error like this "Error: Internal profiling error 3072:999". I am able to profile with 18 stream though.
Thanks,
Ray
Hi!
I’ve the same problem. Only with a different code:
Internal profiler error b[/b]
Also, I use this command:
nvprof --analysis-metrics -o analysis.nvprof ./a.out --benchmark -numdevices=2 -i=1 --print-gpu-trace
When I leave out the –analytics-metrics , everything goes fine.
Problem solved! I used the new 7.5 nvprof version.
It is now returning a human readable error (timeout error).