Error: Internal profiling error 4183:7

Hello everyone,

I was trying to profile my application on a Kepler K20Xm, on a debian 10 image. I have installed cuda 1O.0 .130 and driver version 418.113. However, I get the error: Error: Internal profiling error 4183:7. I was able to profile on a fermi GPU M2075, with cuda 8.0 and driver version 367.48. Can anyone relate or explain this type of error. I couldn’t find an answer in the net.

Thank you for your help,
Dorra

Hi,

Are you trying to profile events/metrics using nvprof? Can you pls provide the nvprof options used to profile the application?

This could be related to profiling a specific event/metric or set of events/metrics. Profiling events/metrics related to L2 cache could be the potential reason. Would it be possible for you to give a try to the CUDA toolkit 10.1?

Hi,

I was trying to profile metrics using nvprof. I used backprop from rodinia benchmark
command: nvprof --metrics all ./backprop $inputsize

I tried with cuda 10.1.243, and I get this error:
./backprop: error while loading shared libraries: libcudart.so.10.0: cannot open shared object file: No such file or directory

I updated my variable LD_LIBRARY_PATH, but still got that error.

Thanks for your return,
Dorra

Hi Dorra,

Can you please profile only those metrics which are of interest? I highly suspect issue is due to the profiling of L2 cache metrics; nvprof option ‘–metrics all’ profiles all available metrics.

Hello,

I tried with some metrics, but still the same issue (with cuda 10.0.130 ). I even tried on another kernel from cuda SDK samples: matrixMulCUBLAS; now I get this error by running: nvprof -m ipc ./matrixMulCUBLAS –profiling-semaphore-pool-size 100000
==12561== Error: Internal profiling error 4292:1.