Error when using Nvidia GPU and likwid perfctr

[Srijan Arora]

Hello. I am trying to run a simple matrix multiplication benchmark on a server, and I am constantly encountering the following error. 

--------------------------------------------------------------------------------
CPU name:       Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz
CPU type:       Intel Skylake processor
CPU clock:      3.41 GHz
--------------------------------------------------------------------------------
Error setting up GPU Marker API.
--------------------------------------------------------------------------------
GPU Marker API result file does not exist. This may happen if the application has not called LIKWID_GPUMARKER_CLOSE.

For reference to my hardware, please find screenshots of the system I am working on attached as a screenshot.

I am trying to compile the program named benchmark1.cu, (attached below) using the following command:

/usr/local/cuda-11.0/bin/nvcc -I. -Xcompiler -I/usr/local/include -DLIKWID_PERFMON -DLIKWID_NVMON benchmark1.cu -o b1 -lm -L/usr/local/lib -llikwid

Running the code with the following command:

likwid-perfctr -W FLOPS_DP -m ./b1

causes the issue mentioned above. Please see if you can find some way to resolve this. 

Thanks and regards

[Thomas Gruber]

Hi,

which CUDA version are you using? Can you please specify the GPU ID on the command line and try again:
likwid-perfctr -G 0 -W FLOPS_DP -V 3 -m ./b1

Please capture the stdout and stderr and provide the output.

I can run your benchmark just fine using your commands.

Best,
Thomas