High CPU usage while running CUDA programs

Hi all,

CUDA was working fine on my linux system until a couple of days back. Suddenly, whenever I run any CUDA program, the CPU usage shoots very high (sometimes to 350%). I have no idea why this happens. Even the bandWidthTest program takes forever to complete.

I am using a Tesla card on a redhat linux 4.x system. There is nothing wrong with the system as such as other applications/programs run fine.

Any help will be appreciated. Thanks!

Does this only happen when you run programs you’ve compiled yourself, or with any program at all (e.g. the included binaries in the CUDA SDK)? Also, have you made any changes to the device drivers for the system (even if it was done automatically through a system update, etc.)?

I am not sure if there were binaries which came along with the SDK. But I compiled the (unmodified) source code in the SDK (bandwidthTest for eg.) and when executing the binary that is generated I see this problem.

I have not made any changes to the device driver. Would you suggest re-installing the device driver?


Please generate and attach an nvidia-bug-report.log. Also, please post the actual output from the apps.

Log attached. Please remove the .txt extension after you download.
nvidia_bug_report.log.txt (202 KB)

Your system seems to be experiencing a number of problems. First, kde is segfaulting. Also, 177.13 is no longer supported (and hasn’t been since August).

You should upgrade to the CUDA_2.0 release, and eliminate the OS based instability.

Additionally, you didn’t provide the actual app output.


Thanks for the reply. I just confirmed with my team mate who is using the same server that “kde” is actually name of his program and not the linux kde. Also, I do not have any output since the program does not run to completion. The bandwidthTest program starts and hangs in the middle (while copying data) for a long time. I shall try to re-install the driver and post the result. Meanwhile, any other suggestions are welcome.