Kenel or display Timeout on non Display GPU under Windows Can it?

Accelerated Computing CUDA CUDA Programming and Performance
1

Is there any way a kernel can time out under Windows 7/64 on a non-display GPU? This appears to be possibly what is happening, but it is hard to tell. I’m doing a long running kernel<<<1,1,>>>() call in order to wire up my neural net serially. If I have minimal connectivity between neurons (4 synapses per neuron), the kernel completes and the program works as designed. If I increase connectivity to 12 synapses per neuron, the kernel fails perhaps, or perhaps the driver is timing out. I see an error after a few seconds of kernel time “Display driver stopped responding and recovered”, and I get no errors in my program, but the neural net display is no longer displaying although things are still running. None of my kernel calls set an error during this.

So, I suppose there are two questions:

  1. Can a non-display GPU (Tesla C1060) kernel time out?
  2. Can the display GPU (Quadro 3700) time out while waiting for the non-display GPU kernel to complete? And perhaps lose my display context?

Thanks,
Ken Chaffin

2

As far as i know on windows, if the system recognizes the hardware as a display driver then the watchdog will time out on it even if there is no display attached. you can eliminate or postpone the time out with a little registry edit.