Card used 4 display - How to find it? Multi-GPU code

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I would like to know if there is an API through which I can find if the CUDA device is being used for display or not.

I am into some serious multi-GPU stuff and this would help.

Thanks for any help!

Best Regards,
Sarnath

2

If you look at the forum post with the release of the CUDA 2.1 beta, Tim (I think it was Tim) said that there is a feature in 2.1 final that didn’t make it into 2.1 beta: cudaGetDeviceProperties will have a flag telling you if there is a watchdog timer enabled on that GPU or not. That isn’t quite the same as asking if it is attached to a display, but might be what you really want.

3

Yeah, you could use the watchdog timer query, but it’s not necessarily always what you want. Can you give a better description of what you’re actually trying to do?

4

Thank you Tim and MrAnderson.

I want to spawn a multi-GPU kernel that is going to run for minutes together (one single invocation…). So, I dont want a watchdog monitored GPU for this case. Thanks.

5

Okay, so you actually want watchdog timer query. Wait for 2.1 final! Coming soon…

6

Thanks. Looking forward for it.

Until then, I would run a 30-sec dummy kernel to figure out if there is a watchdog or not :-)