Earlier this spring, New York State and six academic partners, including the State University of New York (SUNY), committed to create Empire AI*, a GPU supercluster for AI and high-performance computing academic research. Stony Brook faculty member Dr. Robert Harrison is serving as the interim director of Empire AI and can be a useful resource to faculty interested in exploring this resource. An early version of Empire AI’s research computing instrument, called “Alpha,” is coming online by early November 2024. The hardware specifications for Alpha are outlined at the end of this email message. While the ultimate Empire AI instrument will far outstrip Alpha’s specifications, our research communities can now start to use this shared resource. Alpha, should enable Stony Brook to do research we otherwise could not with existing resources.
- HOW: Please submit brief work order requests (WORs) to Empire AI via this secure online form. This short form asks for a paragraph description of your research project and five questions about your resource needs.
- WHEN: You may submit your work order request starting October 3 through October 31, 2024. When you submit WORs within this time window has no bearing on when and whether the project is run on Alpha.
- WHAT: Research compute jobs can be of any CPU/GPU-scale and duration, given the hardware specifications, guidelines, and context below.
- WHY EARLY ADOPTERS: Alpha will be in start-up mode during its initial months of operation, meaning early users will be expected to help work out configuration issues, and ensure the necessary software is installed.
- Empire AI will allocate resources for this initial phase of Alpha following these operating principles: To achieve equity of usage across the six academic systems sharing the instrument, and to maximize utilization of Alpha’s capacity.
- Priority will be given to projects that best harness the capabilities of Alpha.
- Depending on the total number of work orders submitted by any one academic institution and the distribution of work orders submitted among institutions, Empire AI may ask a university to prioritize its corresponding work order requests, following whatever procedure that academic institution chooses.
- When considering appropriate research data for use of Alpha, note that Alpha is neither HIPAA nor NIST-800-171 compliant.
- Empire AI expects to make Alpha available in this fashion for about one year (i.e., through approximately November 2025), subject to change with notice to the user community.
Finally, it is through the generosity of the Simons Foundation and its Flatiron Institute that Empire AI can so rapidly provide access to an initial instrument for our shared research use. Through these means, there are no institutional or user fees for running compute jobs on Alpha. We are piloting Alpha to understand user demand, usage patterns, classes of users, and types of jobs and workloads. Empire AI and Stony Brook University will use our collective experience with Alpha to plan user fees and resource allocation for future builds of Empire AI.
If you have any questions, please contact Robert Harrison at Robert.Harrison@stonybrook.edu
Please consider submitting a work order to take advantage of this opportunity. I sincerely hope that Stony Brook alone overwhelms the Alpha system, but in any case it will help us understand the need for this and additional GPU resources.
Vice President for Research
Stony Brook University
kevin.gardner@stonybrook.edu
O: 631.632.7932 | M: 603.767.
Empire AI: Alpha Hardware Specifications
12 HGX Nodes
- 8 H100 80GB GPUs
- 10 400Gb/s ConnectX-7 NIC Cards (8 for IB and 2 for Ethernet)
- 30TB NVMe caching space
- 2TB of system memory
- 8 network switches, 96 optical connections
- 2 login nodes and 2 cluster management nodes (NVIDIA Base Command with licenses for all gear)
- 4 x 720TB Flash storage (home directories, training data, snap shots)