When the execution GPU is also driving a display, under Windows or Linux, the CUDA runtime terminates kernels that run for more than a certain amount of time. Is there a way to force the same behavior for a headless machine (ie, one that does not run X)? Alternatively, is there any way to get the runtime to kill a kernel on command, without forcing a full reboot?
Not that I know of unfortunately…
On linux Ctrl+C usually gets things done for me, without reboot. Not sure if that’s what you need.
That does not work when there is some infinite loop running on GPU for me. Restarting X results in a machine hang, even shutting down hangs the machine when trying to shut down X.