IBM Spectrum Computing Group

Hi Zheng Fa, >When I try to submit a pytorch ddp job with 16 GPUs on 2 nodes, I find it doesn't work. #BSUB -gpu "mode=exclusive_process:aff=yes" #BSUB -R "32*{rusage[ngpus_physical=8]}+32*{rusage[ngpus_physical=8]}" Replace the above ...

torchrun needs to be executed on all hosts (as you found out), so we use something like blaunch -z $HOSTLIST --nnodes=2 --nproc_per_node=8 --rdzv_id=100 --rdzv_backend=c10d --rdzv_endpoint=$HOSTNAME:29400 ./elastic_ddp.py where $HOSTLIST contains ...

Hi， I am a user of IBM Spectrum LSF 10.1. When I try to submit a pytorch ddp job with 16 GPUs on 2 nodes, I find it doesn't work. LSF will allocate 2 nodes with 16 GPUs to the job, but the job doesn't run correctly. It will block until I run the ...

The materials from the LSF User Group on May 16th in Santa Clara, are now available in the Library Thanks to all who attended. ------------------------------ Bill McMillan Principal Product Manager, IBM Spectrum Computing ------------------------ ...

More can be found in quick stat guide . ------------------------------ YI SUN ------------------------------

Hi Nitin, You will need to review and accept the license agreement . There are some cluster scale limitations in this edition, e.g. number of nodes, max number of cores on each node, max number of jobs to run, etc. You may not be able to use some advanced ...