Bridges Phase 2 Early User Information

 

Please check this document frequently for updates.

 

Charging

You will be charged for the use of Bridges' Phase 2 nodes.  However, these charges will not be recorded with XSEDE until the Phase 2 hardware is officially accepted by the NSF and declared in production.  Until then, charges will accumulate and you can track them by using the projects command on Bridges.  These charges will not be reflected in the XSEDE portal until the Phase 2 hardware is in production.

 Quick notes

  • Because the Phase 2 GPU nodes have P100 GPUs, not the K80 GPUs in the Phase 1 GPU nodes, you should recompile your GPU codes.

    Use this flag on the nvcc command:

    -gencode=arch=compute_60,code=sm_60
  • The debug partitions are to be used for short debug or benchmarking runs only.  Please do not run prodution jobs in the debug partitions.

 

Partitions

There are three partitions available for Phase 2 Early use.   They have been configured to provide equitable access for all Phase 2 users and are subject to change to better serve user needs.  If you have comments about the Phase 2 partitions or find that they do not meet your needs, please email bridges@psc.edu.

The partitions are:

  • GPU_100 - The nodes in this partition contain 2 NVIDIA P100 GPUs along with 2 Intel Xeon E5 series CPUs. You should recompile your code for all jobs in this partition.  See the Quick notes list for the option to use on the compile command.
  • GPU_100-debug - This partition is for short debug or benchmarking runs.  Do not run production jobs in this partition.

    The nodes in this partition contain 2 NVIDIA P100 GPUs along with 2 Intel Xeon E5 series CPUs. You should recompile your code for all jobs in this partition.  See the Quick notes list for the option to use on the compile command.

  • LM_100 - This partition contains 34 LSM nodes with 4 Intel Xeon E7 series CPUs and 3TB RAM each, and 2 ESM nodes with 16 Intel Xeon E7 series CPUs and 12TB RAM each.

 

Submit jobs to the GPU_100 and LM_100 partitions using the sbatch command.

Request resources in the GPU_100-debug partition using the interact command.

 

For more information on the sbatch and interact commands and running jobs on Bridges, see the Running Jobs section of the Bridges User Guide at https://www.psc.edu/index.php/bridges/users/user-guide

 

 Partition limits

PartitionDefault # nodesMax. # nodesDefault timeMaximum time
GPU_100 1 4 30 minutes 1 days
GPU_100-debug 1 1 30 minutes 1 hour
LM_100 1 1 30 minutes 4 days