Available CPU count on node with CR_ONE_TASK_PER_CORE
If CR_ONE_TASK_PER_CORE is configured then the core count rather than the CPU count of a node is used to determine if a node can be used by a job. This can result in a job being rejected than should be able to run. Sample configuration and job below: SelectTypeParameters=CR_Core_Memory,CR_CORE_DEFAULT_DIST_BLOCK,CR_ONE_TASK_PER_CORE NodeName=psg-dgx2-01 NodeAddr=jette NodeHostName=jette RealMemory=1536000 Gres=gpu:16 Sockets=2 CoresPerSocket=24 ThreadsPerCore=2 State=UNKNOWN $ srun --gpus-per-task=1 -n1 --cpus-per-gpu=64 -J test39.7 -t1 ./test39.7.input srun: error: CPU count per node can not be satisfied srun: error: Unable to allocate resources: Requested node configuration is not available
Please register or sign in to comment