Kempner Institute Partitions

Overview

For questions regarding usage and etiquette of the Kempner resources or the “Kempner Institute Handbook“, please contact the Kempner Support. You can also ping their #cluster-users channel in the Kempner slack space with Kempner-cluster specific questions/concerns, and advice. For access to the slack channel please contact Nikki Shawn.

NOTES: Recent changes mean that not all members of a Kempner lab can run jobs. You must first request access to run jobs. Please consult your PI for details on gaining Kempner job privileges. Kempner expects its users to review the Kempner Institute HPC Handbook before using the cluster.

The Kempner Institute for the Study of Natural and Artificial Intelligence has their own purchased partition on the FASRC cluster.  These partitions are open to users affiliated with Kempner and their allocation is governed by the relative fairshare of each group. If yours is a Kempner-affiliated lab group and you have been given explicit access to run jobs in Kempner you can submit to the partition just as you would any other partition but you must indicate which fairshare pool you are using by adding #SBATCH --account=<kempner_fairshare_pool>.

A dashboard showing the relative allocations can be found here (RC VPN or Harvard network required).  The partition is the following:

• kempner: This block of GPU’s contains 2304 Intel Ice Lake cores and 144 Nvidia A100 GPUs. Each node is water-cooled and contains 64 cores, 1TB of RAM, and 4 GPU’s. The nodes are interconnected with HDR Infiniband. This partition has a 7 day time limit.
• kempner_h100: This block of GPU’s contains 9216 AMD Genoa cores and 384 Nvidia H100 GPUs. Each node is water-cooled and contains 96 cores, 1.5TB of RAM, and 4 GPU’s. The nodes are interconnected with NDR Infiniband. This partition has a 3 day time limit.
• kempner_requeue: This partition underlays the kempner and kempner_h100 partitions and is intended for work that can be requeued by higher priority work. This behavior is the same as the public requeue partitions of gpu_requeue and serial_requeue. Also similar to the public requeue partitions this partition is half cost with respect to fairshare. This partition has a 7 day time limit.

For more information about Slurm partitions on the FAS RC cluster, please refer to the Running Jobs document.  For information for how to run jobs on GPU’s on the rest of the Cannon cluster that are not in Kempner see this document.

PLEASE NOTE – HMS/Kempner: If you are one of the Kempner groups whose PI is based out of HMS, you will not be able to submit ordinary jobs to the FASRC cluster. If you mistakenly submit a job to one of our normal partitions, the job will pend forever. Please contact Spencer Pruitt at HMS if you are a Kempner HMS user who needs access.

Access

Access to the partition is via membership in a existing kempner lab group on the FASRC cluster and approval after filling out a Kempner access form. Please consult your PI for instructions and a link to the form. This process is managed by Kempner, FASRC staff cannot add you to Kempner resources directly.

• For PIs who believe their lab should belong to kempner_users,  please contact Max Shad or Elise Porter at the Kempner Institute.
• For individuals who believe they should have access, please consult your PI for instructions and a link to the Kempner access form or contact Kempner Support

Fairshare

FASRC’s general fairshare policy applies to the Kempner cluster. However in the case of Kempner cluster the fairshare is split from your normal lab fairshare into a specific pool for your lab or group.

• kempner_<pi_lab>: Is the lab specific fairshare pool. Similar to the normal FASRC fairshare policy all users in these pools pull from the same common allocation block.
• kempner_grads: This pool is for graduate students who are affiliated with Kempner but may or may not be members of Kempner affiliated labs. Each user in this pool has their own fairshare allocation, however there is a cap to the common fairshare of based off of the full group usage.
• kempner_fellows: This pool is for postdocs and other fellows who are affiliated with Kempner but may or may not be members of Kempner affiliated labs. Each user in this pool has their own fairshare allocation, however there is a cap to the common fairshare of based off of the full group usage.
• kempner_hms: This pool is for researchers from HMS who are affiliated with Kempner but do not have sponsorship via a lab.  Each user in this pool has their own fairshare allocation, however there is a cap to the common fairshare of based off of the full group usage. Spencer Pruitt from HMS manages access to this group. Please contact him for access questions.
• kempner_undergrads: This pool is for undergraduate students who are affiliated with Kempner but may or may not be members of Kempner affiliated labs. Each user in this pool has their own fairshare allocation, however there is a cap to the common fairshare of based off of the full group usage.

By default users on FASRC resources utilize their primary group for fairshare, this also applies to Kempner. Therefore users using Kempner resources should include the --account flag and indicate what pool they wish to use when running jobs on the Kempner cluster. Kempner fairshare accounts are not to be used outside of the Kempner partitions, for other resources on Cannon users should use their normal fairshare accounts.

Contact Information

To contact specific individuals at the Kempner Institute, please browse to their contact page.