Protip: really old GPUs appear to block NSight OpenGL tracing

I was puzzled all morning over why Nsight’s “Start Performance Analysis” wasn’t capturing any OpenGL traces and then cryptically reporting that there was an error.

Disabling the ancient 9400 GT hiding in my system and restarting VS2012 solved the problem:

I disabled the GT 240 (sm_12) as well, just to be safe. :)

Both of the ancient GPUs and the 750 Ti are running headless.

The win here is that I didn’t have to open the case and pull the card since I still like to see how CUDA code runs on old GPU architectures.

A warning message from Nsight might be appropriate as I had to guess that this was the issue.

Hi Allan,

Sorry for the delay. Yes, we did stop support for pre-Fermi GPUs, however, you should have not gotten any data for the pre-Fermi GPUs, but if the context was running on the Fermi and above GPUs, you should have gotten data for them.

Was the app running on the pre-Fermi GPU, only Fermi and better or a mix?

PS: I am referring to “pre-Fermi” to avoid the confusion between Tesla arch and Tesla product :-)

The profiled app was running solely on the GTX 680 sm_30 Kepler device. No data was returned.

Furthermore, when my machine comes out of S3 suspend with the disabled cards… it reboots because those two cards are disabled. I just reenabled the disabled cards and the machine resumes properly.

Sorry, I’m a bug magnet! :)