I have two GPUs in my machine - a GeForce GTX470 (with a display connected) and Tesla C2050 on openSUSE 11.4. I’m running some CUDA code on the Tesla C2050 however if I want to terminate it or issue a kill command to it, it locks up the entire graphical output of the machine and there is a huge amount of linux kernel use on the CPU (Core i7).
I’m guessing this is because the OS tries to kill all the threads on the GPU by issuing a command to each of them. It really doesn’t seem like the best way to do it with such a huge number of threads…
The main things I want to know is:
- Is the correct way to kill this code? Is there something in CUDA which can be issued to kill it instead?
- Has anyone else experienced this before?
Any help in answering this would be appreciated.