After CUDA debugging ends, gpu remains in use

When I am done debugging and the process ends successfully, the next program to request a GPU will wait forever.

When I check nvidia-smi, it tells me the device is 99% utilized.

Anybody any clue? Maybe some stray process which doesn’t get killed? Known bug in nsight? Something else?

My environment:
Windows 7 ultimate, SP1
Visual studio 2012, update 4
CUDA 5.5
Nsight 3.2.2

Ps. the only solution now is to release the devices via the device manager, after which usually the machine crashes…


Hi ikku100,
it looks that you are using an old version. could you try with Latest version: CUDA 6.5 and Nsight 4.1?