Kernel timeout on GTX680, but works fine on GTX480


I got two linux machines with GTX680 and GTX480 installed respectively. The same kernel, which works fine on GTX480, encounters “cuda error code=6 cudaErrorLaunchTimeout”. Anyone has any idea of what’s happening on GTX680? Does GTX680 has a lower threshold of time limit for the kernel? How can I run the kernel that might takes more than several seconds?


Oh, by the way, I checked the devProp.kernelExecTimeoutEnabled and it is true. How to set this device property to false?

That property is set when an X server is running on the GPU. If you can drop to console-only mode (how to do this depends on your Linux distribution), then you will be able to run CUDA programs without a watchdog timer on either card.