I have a few kernels which exceed the windows watchdog timer limit if the input data is too large.
In every cuda document i read that a device with no display attached to it is not affected by the watchdog timer.
I have a Geforce 7 as my main device and a Geforce GTX 480 as my cuda device with no display attached.
The watchdog timer still monitors the Geforce GTX card and terminates the kernel, how can this be possible?
So i used the registry hack to set my watchdog timeout to a larger value and now i can process larger input data.
But if my input data gets too large, my program freezes and i have to shut the process down with the taskmanager.
Can this happen if the watchdog timeout is set too large?, i set it to 512 for testing purpose, or my algorithm has some errors.
Anyway, the geforce gtx 480 should never be affected by it and thats my main question!