CUDA Profiler API Trace on command line


I would like to do a CUDA API Trace using the NVIDIA Visual Profiler.

However, I have to run my application on a remote machine, which I can only access via SSH. So I need to run the Visual Profiler on command line, using the environment variables COMPUTE_PROFILE=1, etc.

Now I would like to do a CUDA API Trace. But I could not get that working. Are there any environment variables to set or are there other things to do to get this working?

I appreciate your help!