Cuda known limitations

In the developer release notes for Cuda 1.0 under Linux it is stated “Individual GPU program launches are limited to a run time of less than 5 seconds on a GPU with a display attached.” I believe this limitation also exists for Windows.

Does anyone know if it will be removed in a future release? (or should we just stick to using GLSL) <img src=‘http://hqnveipbwb20/public/style_emoticons/<#EMO_DIR#>/crying.gif’ class=‘bbc_emoticon’ alt=‘:’(’ />

This is an OS limitation, not a CUDA limitation.

Since the 5 second limitation is the watchdog timer built into windows, I don’t think even GLSL can get around it, or am I wrong?

Maybe you’re confused by the choice of language: A “GPU Program” launch is a single kernel running on the GPU. A CUDA program can run for as long as it wants, as long as each individual kernel call is less than 5s.

The timeout is to stop your system hanging if the graphics driver loops.

In Windows Vista the limit is reduced to 2 seconds. Microsoft documents registry hacks to get around it. But note that this means looping graphics code can hang your system.

The same issue exists in Linux according to the Cuda 2.2 development kit release notes http://developer.download.nvidia.com/compu…notes_linux.txt

I have no idea how to change or remove the limit in Linux.

The timeout is to stop your system hanging if the graphics driver loops.

In Windows Vista the limit is reduced to 2 seconds. Microsoft documents registry hacks to get around it. But note that this means looping graphics code can hang your system.

The same issue exists in Linux according to the Cuda 2.2 development kit release notes http://developer.download.nvidia.com/compu…notes_linux.txt

I have no idea how to change or remove the limit in Linux.