NSight causes remotely debugged app to peg CPU at 100% and lock up?

Problem: NSight appears to behave correctly right up to when “Pause and Capture” is clicked at which point the interactive hud on the remote machine locks, and the process for the app being tested goes to 100% cpu utilization (2 core machine). The diagnostic dump appears correctly in the attached instance of Visual Studio 2009, but is very slow to respond and typically crashes within 10-20 interactions with the diagnostic panel (this is highly irregular however, sometimes allowing particular features, while immediately locking Visual Studio other times).

Additional info:
-The app being tested is DX9
-Both DX9 and DX10 Nvidia demo apps run and can be profiled fine using the same machine/network configuration.
-The dev machine is a core i7, 240GT, 18GB ram, the remote is a Core2 duo 2.4, 9800GT, 6GB
-Ctrl-Z on the remote brings up the HUD as expected, and until “Pause and Capture” is pressed appears to function correctly, but then locks immediately (hourglass, 100% cpu utilization).
-running debug Directx isn’t producing any errors we’ve seen.
-performance markers appear to be correctly implemented, and show an appropriate hierarchy of queues/calls in the analysis tool compared to how the code has been annotated.
-selecting “Resume from Capture” in the VS Nsight menu usually works, and returns control of the app to the remote machine.

The Question: Given that NVidia’s demo apps profile without any issues, are there any “commmon” problems we can look out for that would cause the aforementioned symptoms when trying to profile our own app?

Hi tcoles,

Which version of Nsight are you using, and which GPU driver version do you have installed on the target machine?


NSight is version 2.2 (build

Driver is 306.23 although I’ve also tried 306.97 and 301.42 which yield the same results.

Could you try the latest build posted (
If it still reproduces, would it be able for us to get a hold of the build (just need to able to run it, so binaries would be fine).