I’ve just started developing CUDA applications on my laptop for work, since preliminary benchmarks seem so promising (10x-3000x performance? Yes please!) but I’ve run into a small snag: CUDA seems to cause Vista some headaches.
I’ll be trying to install Windows XP and see if I can run it there instead, but the problem I’m encountering is that when CUDA code of any computational complexity tries to run, about 1 second in, Vista halts and reboots the display driver, thus aborting the CUDA program. This is a real problem since I was planning on eventually running a CUDA app over night and this is going to get in the way.
I suppose my question is: is there any way around this? Either with another OS (Does Linux get around this somehow? Windows XP?), or some modification somewhere. Changing a setting so that Windows stops caring if a display driver doesn’t respond. Even upping the delay from what seems to be 1 second to maybe a minute, or 10 minutes (Or longer if I intend the code to run for longer). Is there something I can add in the code that lets Windows interject in there to check responsiveness… Any solution is an option really (See caveat in next sentence).
Keep in mind, I’m on a laptop, and so adding in another video adapter isn’t an option.