It is pretty weird actually, since my programs are not even close to being as intense as the sdk samples… so I guess it is not a matter of resources (which I thought that was the case in the beginning)
and while the sdk samples run just fine without needing to kill gnome, my programs execute, but they give no results (actually all the elements of the matrices I calculate turn to be something like 2,39 * 10^-38). If I switch to a tty (without killing gdm), then my programs work like a charm.
any ideas? The laptop all this happens on has a 9300m GS gpu …
thanks in advance
The issue is probably related to the watchdog timer. When the GUI is running, a kernel that takes more than a few seconds will be terminated before completion.
Are you checking the return codes from each CUDA call? If your kernel is terminated, you should get a return code != cudaSuccess on the next call after it finishes running (most likely the next cudaMemcpy).
thank you very much for responding…
I’ve read about the watchdog timer… and it is supposed to limit the time dedicated to a single kernel call to about 5 to 6 seconds. In my case in 12 seconds I make about 3.000 kernel calls with a for loop. So logically it can’t be it. I’ll check to see the return code though…