I’m using Vista 64, but I think this problem may affect other developers as well. I’ve noticed that after a bad memory access in the device memory space, my screen tends to get a lot of random snow that won’t go away unless I reboot.
Is it possible to reload the driver or force Windows Vista to use a more generic video adapter driver that won’t cause this to happen? It looks like the device doesn’t offer memory protection, and so it’s possible for a CUDA application to scribble onto memory areas that are in use by the display driver. While this could have potential benefits, for now it’s a major pitfall to dereference the wrong pointer and end up corrupting your display device.
Can someone from NVidia confirm this or provide a workaround?