5 seconds kernel runtime limit on CUDA 2.1

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, I’m new to CUDA, so please bear with me on this. Thanks. I’m planning to run CUDA on GeForce 9800GT on a Dell PC, which does not support the option of connecting the display to the onboard integrated graphics adaptor. So, it seems the only option on this machine is to connect the display to 9800GT at the same time of CUDA computing.

The question is: does CUDA 2.1 beta still have the 5 seconds kernel runtime limit on a graphics card attached to a display? Or, is there any other solution to disattach the 9800GT from the display in this case? Thanks.

2

As far as i know its still there and probably will not get solved any time soon since its something the windows manager does (in this case, windows), not CUDA.

3

Thanks, Ailleur. So, do window managers in all Linux distribtuions have the same issue? which means they are also subject to this 5 second limit on linux?

4

Yes, but you can decide to not run X.org on Linux, and avoid the watchdog timer that way.

5

That’s a good idea, thanks. I’ll try it on linux.

6

In theory you can use the registry hack from the release notes to disable the watchdog on Vista, but I have no idea how long you can leave it off. I know we’ve done 9 seconds in house and not had a problem, but maybe you can go higher.