Slurm Cluster

AI Compute Resource (AICR) AI Compute Resource (AICR)

A cluster intended for large-scale, high-vram GPU jobs. Operated by the MGHPCC.

Beta
AI Compute Resource (AICR) Features
GPU Compute
High VRAM GPUs
Parallel GPU Compute
Batch Compute
System Overview
50
GPU Nodes
400
GPUs
32
Service Nodes
~14 PiB
All-Flash Storage
AICR is a GPU-focused AI computing cluster operated by the MGHPCC.It runs NVIDIA B200 and RTX PRO 6000 GPUs behind a Slurm scheduler with InfiniBand NDR400 interconnect and VAST Data all-flash storage, and is optimized for large-scale AI/ML training and other high-VRAM GPU workloads. See the AICR system description for details.

Use cases

AICR is a high-powered, large-VRAM GPU cluster which is intended for large scale GPU compute. If you need to leverage many GPUs towards model training or fine tuning, AICR is the right system.

warning
AICR is not an appropirate system for hosting models for interactive inference (i.e., chatbot or coding agent that waits for your input before proceeding). If you use AICR for this purpose, your account will be removed. Inference that is fully automated in a batch job and makes approporate use of the GPUs is acceptable.

Getting access

AICR access is by proposal only. In order to be condsidered, gathering the following information:

  1. Verify that your code can scale across GPUs efficiently. Not sure if it can? Reach out for help on how to run scaling studies on Unity.
  2. Estimate the number of B200 GPU and RTX Pro 6000 GPU hours your project needs. Unlike Unity, AICR is allocated on a per-project basis with a use cap.
  3. Submit the AICR project request form. Proposals are evaluated after the first and third Thursday each month.
stylus_note
To submit an AICR request form and use the platform, you must be registered through the Unity Platform account portal. You don’t need to be in a Unity cluster PI group. To initialize your account, simply log in to the account portal without proceeding to request Unity group access.

Getting help

The AICR system has a separate documentation site. For proposal and AICR help, contact the Unity team.