ORCD Newsletter: January 2025

Above the Fold - New Hardware for ORCD

In the coming months, ORCD will install new hardware in MGHPCC for use by MIT Researchers.

• Working with the Deans and Provost, ORCD has ordered 276 new GPUs for general use by MIT researchers. MIT placed the $6.5M order two weeks ago. Delivery will happen in February, and we expect the machines to run in early March. We plan to make them available for MIT researchers in April.
• Using Project Olympia funds, ORCD plans a second purchase of 60 GPUs later in the year, after we gain experience with the first installation. 
• Following Gov. Healy’s announcement in December, a sub-committee of the MGHPCC board and others have begun working on the design of an AI system for the State AI Hub, eventually costing $120M ($10M from each of the six MGHPCC partners and $60M from the State in two $30M tranches). On average, this will add about 96 GPUs this year.

Putting this all together, ORCD will have around 392 new GPUs in service by the end of CY2025. The table below shows how the ORCD GPU fleet is currently planned to grow under these commitments, totaling $25M:

We plan to make these machines available at no cost in CY2025 and then in CY2026, move to a model with an essential no-cost availability for all researchers, an MIT-wide fee for service and spot/rental market, and a reserved portion allocated by the Schools. ORCD, the VPR, and others are working on how this will work.

First ORCD Seed Fund

Through Project Olympia support, ORCD is launching its first ever seed fund round. Information about the seed fund opportunity can be found at https://orcd.mit.edu/seed_fund/spring2025.

This round of funds is targeting projects that need short-term support and some enhanced GPU access to achieve a tangible research outcome such as a publication, thesis chapter, or conference submission. We are looking for projects that will also provide feedback to the ORCD team on services and areas to prioritize in the future. Solicitation responses are due February 28th. We expect to support about six projects in our first cycle.

What We’re Reading

• Google has finally demonstrated scaling in error correction: the performance improves proportionally as you add qubits for error correction. Many say this is HUGE.
• Getting hit by space debris has gone from a punch line to a serious problem.
• LLMs running out of training data has become a thing.  
• Harvard students used Meta’s smart glasses and AI to get private info from people’s photos. What can go wrong with that?
• Timely: MIT to become Silicon Valley of the Aging World and Peter just turned 65.
• Vannevar Bush on the origins of the modern university research ecosystem, cost recovery, and much more: Pieces of the Action.