Describe the bug
When using more than 1 GPU; if you hit C to create a checkpoint and wait until 1 GPU isn't being used but 1 GPU is and then you hit C again (to cancel the checkpoint); it does not resume work for the GPU's that were no longer being used

Expected behavior
When using more than 1 GPU; if you hit C to create a checkpoint and wait until 1 GPU isn't being used but 1 GPU is and then you hit C again (to cancel the checkpoint); it should resume work on ALL GPU's original assigned to the run