5 seconds running time limitation

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello, I see in CUDA release notes about a 5 seconds running time limitation. Does that mean a CUDA program cannot be too long to finish one time execution within 5 seconds? I have tried CUDA examples on a GPU attached to X and didn’t see any limitation about running time.

Also the CUDA release notes says X must be initialized at least once. How can I do this to work with the 5 seconds limitation? My machine has one GPU only. Should I boot my machine, start X, then stop X, and then finally able to run CUDA program and doing this again and again whenever I have to run a CUDA program??? Weird…

Many thanks for helping me out…

2

That’s 5 (or so) seconds per a kernel. If there are many kernels launched one after another (as is most often the case) than the program can run indefinitely as long as each iteration finishes fast enough. Usually kernels take in the order of milliseconds.

3

You need an X configuration to be able to install the NVidia driver. However once the driver is installed you can run the mashine completely without X. You just have to make sure that the nvidia module is loaded.