I have system with 2 GPUs; a GeForce GTX560 and a GeForce 210 (CUDA toolkit 4.1 and nsight 2.1). When I tried to run CUDA debugging I've got this:
So, how can I change which card I use with the debugger?
I have system with 2 GPUs; a GeForce GTX560 and a GeForce 210 (CUDA toolkit 4.1 and nsight 2.1). When I tried to run CUDA debugging I've got this:
So, how can I change which card I use with the debugger?
Here are two options:
1) Add support in your program for specifying the target device. If you add a command line option you can set the value in the Nexus Properties dialog available through the Solution Explorer Project node context menu.
2) Use the environment variable CUDA_SET_VISIBLE_DEVICES=1,0 to reorder your two devices. Nsight 2.1 does not support specifying environment variables through the debugger so you have to do the following steps on the target machine (may not be your local machine).
set CUDA_VISIBLE_DEVICES=1,0
Nsight.Monitor.exe
You can also set your global environment variables but you have to make sure you restart the Nsight Monitor so it will pick up the changes. This will affect all CUDA programs so I recommend you limit the changes to just programs launched via the Nsight Monitor.