Using a 2nd gpu device with USER-CUDA package

Hi everyone,

  My node has a dual-gpu installed. The system recognizes devices 0 and 1.
When I submit the first job on this node, everything runs fine on device
0. As I submit a second job, however, the system allocates it on the
same device 0 instead of going for device 1, which is free. Is this
supposed to be so? If not, how can I tell the system to use a specific

  Best regards,


I had a similar problem recently. This reply by Christian Trott should help you: