I’ve finally managed to setup profiling through Matlab but now it has stopped working. After the profiling has completed I get “Unable to collect metric and event values. org.eclipse.swt.SWTException.: Failed to execute runnable (java.long.OutOfMemoryError: Java heap space)”
I have 16 GB of memory so it should not be a problem, and it worked yesterday. Any ideas? I use CentOS 6.3.
Can you try editing ${CUDA_TOOLKIT}/libnvvp/nvvp.ini and appending following lines:
-vmargs
-Xms40m
-Xmx2G
This should let the Visual Profiler to use up to 2G of memory.
If you have a session file that can reproduce the problem, we would appreciate if you send it to cudatools@nvidia.com - this way we would try to analyze why the profiler needs more memory then anticipated.
Still the same error, should I increase to 4 or 8 GB?
You can try that but it is unlikely that the memory limit would help…
We would really appreciate if you would contact cudatools@nvidia.com and provide more information. Hopefully we will be able to find a workaround and fix this bug in the future release of the CUDA Toolkit.
OK I have sent an email with all the necessary files.
Thank you. We have received your message and will investigate it.