Longwood HPC cluster

Support for AI and data-driven research

The Longwood high-performance compute (HPC) cluster supports HMS researchers working with AI, machine learning, and data-intensive projects. Hosted at the Massachusetts Green High-Performance Computing Center, it supports both Intel (DGX) and ARM (Grace Hopper) architectures in a secure, flexible computing environment.

The cluster provides scalable GPU power, shared and scratch storage options, containerized tools like Singularity, and job scheduling with Slurm.

Access to the Longwood cluster is available through a secure shell (SSH) at: login.dgx.rc.hms.harvard.edu

Access to the Longwood cluster is available through a secure shell (SSH) at: login.dgx.rc.hms.harvard.edu

Access is limited to devices on the HMS wired LAN, HMS Secure wireless, or the HMS VPN. The sign-in server does not require two-factor authentication, as connections must originate from a trusted HMS network.

Specifications

Includes DGX and Grace Hopper nodes with H100 GPUs.

GPU nodes Architecture

Longwood features 64 NVIDIA H100 GPUs across a heterogeneous architecture. The setup includes:

8 NVIDIA DGX nodes, each equipped with 8 GPUs (80GB VRAM per GPU)
2 Grace Hopper nodes, each with 1 GPU (~96GB VRAM) and 400GB+ of shared RAM

This design supports demanding AI and HPC workloads with substantial memory and compute performance.

Cost

Free access for DIA awardees during the project period.

No cost DIA only

Longwood HPC cluster

Support for AI and data-driven research

Eligibility

Security

Access details

Specifications

Cost