Non-GPU tasks keep using `gpu` partition

I am a CARC user with intensive gpu requirements. For now, my jobs need to queue for a long time, such that even squeue --start -j cannot show the expected starting time.

After checking the current job queue, I found that there are a lot of jobs that do not require GPUs keep using the gpu partition. Most of them requests huge amount of CPUs in one node, in that case, even though GPUs are idle, our jobs cannot get allocated because of the storage of CPU cores.

Here are the screenshots of those non-GPU jobs on gpu partition. I cannot appreciate it more if someone could help me understand the situation.

Hi Zhangyu,

There do appear to be some CPU-only jobs that “block” gpus from being used but they are running on P100 nodes which tend to sit idle. Sometimes users want to make benchmark comparisons between gpu and non-gpu runs but I will keep an out out to make sure our resources are utilized effectively.

I see you have one job in queue which appears to request a40s. These jobs should not delay yours but I will reach out to this user and suggest more appropriate resources if necessary.

Let me know if you have any other questions.

Cesar Sul