Display Driver Stopped responding and has recovered?

Can anyone help me with this problem?

Display Driver Stopped responding and has recovered.Ths is the message i got when i was into a loops which had 125,000 iterations and inside the loops the cudamalloc and cudafree operation was performed with a call to kernel function.

Earlier when i did the same operation with a smaller file which had less points(10,000),i didnt get this kind of problem.But when i tried it with a larger file i got this error message and my screen got blank for couple of seconds.Can anyone please suggest any solution to this

Can this harm or damage my display driver in any ways?

Thanks in advance!!!

Yeah, I also got this error. I’m implementing a raytracer on CUDA. If I sample the image with 256 samples per pixel on 400x400, it works fine. But if I’m raising the sample count to 1024 or raising the resolution to 800x800, CUDA crashes and the display driver is being restored.

Can anyone help us?

I don’t believe the display driver is fully functional while you’re running kernels. If Windows has a call pending to the display driver for a long time, the watchdog timer gets triggered and the driver restarts. There’s a way to disable this, but it’s generally a bad idea.

The answer is to break up your operation into meaty chunks.

Wait a second. Are you telling me I can’t write a CUDA program that runs for more than about 2 seconds?

That I’d have to break the algorithm so it re-enters over and over to get along with the rest of the system?

That is insane if true.

I’m encountering the same problem, when my data is large I get ‘driver stopped responding’ and a ‘Microsoft C++ exception: cudaError_enum at memory location’ written to the output window.

You can’t run single CUDA kernel for more than a few seconds on a device running the primary display. To get around this you have to either run the display on another card, break up your kernel calls to be shorter, or run Linux where you can turn off the GUI entirely. A program using CUDA can run forever, as long as the individual kernel calls are short.

Not so true.

I installed a second card for display and so have entirely access to gpu card for computing

but I still have this problem of kernel duration (under Windows XP).

Or is there something more to do ?

do you have the desktop extended onto the compute card? what does the watchdog timer device property return?

watchdog timer device property : in regedit ?