Geek asks and answers

CUDA - nvidia driver crash while running

I run a raytracer in CUDA with N Bounces (each ray will fail N times).

I view results with openGL.

as soon as N is small (1 ~ 4), everything works fine. as soon as I make N large (~ 10) each thread (about 800x1000) should do a lot of calculations, and this is when the screen goes black and then back, with a note that my accident is nvidia.

I searched on the Internet and now I think that the reason for this is some kind of watchdog timer, since I use the same graphics card for my display and my calculations (the computer takes more than 2 seconds, so the driver itself is reset).

Is there a command to make the host (cpu) WAIT for the device (gpu) as long as it takes?

What should I do? im stuck: (

thank

+3
source share
1 answer

Based on your description, you are running Windows Vista or Windows 7. Windows operating systems have a watchdog timer, you guessed it. The watchdog timer applies only to GPUs with displays turned on.

The simplest solution is to run 2 or more GPUs and run CUDA on GPUs without connecting a display.

You can disable the watchdog timer. See this question for more details . However, you should do this with caution and remember that when you have a lot of cores on your main GPU, you will completely abandon your computer (at least you won’t be able to see what it does) until the kernel completes.

+4
source

More articles:


All Articles