Supercomputers/Systems/Kempner AI Cluster
USA · operational · ai training
Kempner AI Cluster
Also known as Kempner Institute AI cluster
Operated by Kempner Institute at MGHPCC Holyoke .
Phase 1 seed dataset, compiled by hand. These rows were built from public operator, laboratory and vendor sources. A mechanical second-reader pass has since fetched every cited source: 1092 of 1153 systems have a readable citation that names them, and 40 are genuinely weakly sourced. Every claim carries its source and a confidence tier. Treat anything below verified as a lead, not a citation.
Measured performance
- Rmax
- 16.29 PFlop/s
- Rpeak
- 20.02 PFlop/s
- Rmax ÷ Rpeak
- 81.4%
- Cores
- 59,904
- Power
- -
- Per watt
- - GF/W
Figures are as last publicly reported for the configuration described below, not a live measurement. Where a system was upgraded in place, the post-upgrade configuration is shown and the earlier one appears in the timeline.
What this machine is made of
One row per supplier relationship. “Supplier at build” is the company that shipped the part at the time; where that company has since been acquired, the parent it rolls up to today is shown beside it. That distinction is what makes ticker-level aggregation possible across a thirty-year dataset.
| Role | Part | Supplier at build | Quantity | Confidence | Source |
|---|---|---|---|---|---|
| Integrator | - Lenovo ThinkSystem SD665-N V3 (H100 partition submitted to TOP500) | Lenovo | - | Reported | top500.orgnews.harvard.edu |
| Accelerator | NVIDIA H100 SXM5 Accelerators CUDA (Hopper GH100) · 16,896C · 1.98 GHz · TSMC 4N · 700 W 384 NVIDIA H100 80GB GPUs (4 per node) added in fall 2023 | NVIDIA | 384 reported | Reported | news.harvard.edukempnerinstitute.harvard.edu |
| Accelerator | NVIDIA A100 SXM4 40GB Accelerators CUDA (Ampere GA100) · 6,912C · 1.41 GHz · 7 nm · 400 W 144 NVIDIA A100 40GB GPUs (4 per node) from the spring 2023 pilot | NVIDIA | 144 reported | Reported | news.harvard.edukempnerinstitute.harvard.edu |
| Accelerator | NVIDIA H200 SXM Accelerators CUDA (Hopper GH100) · 16,896C · 1.98 GHz · TSMC 4N · 700 W NVIDIA H200 141GB GPUs (4 per node); the Kempner cluster page states 424 after the 2026 expansion but other pages differ | NVIDIA | - | Reported | kempnerinstitute.harvard.edu |
| CPU | - AMD EPYC 9454 48C 2.75GHz (H100 partition hosts) | AMD | - | Reported | top500.org |
| Interconnect | NVIDIA InfiniBand HDR Interconnect Fat tree or dragonfly+ · 200 Gb/s per port · About 0.6 microseconds InfiniBand HDR (H100 partition, per TOP500) | NVIDIA | - | Reported | top500.orgnews.harvard.edu |
AI datacenter metrics
Figures beyond an accelerator count, each with who stated it. What can and cannot be compared across AI datacenters.
| Metric | Value | As of | Basis | Note | Source |
|---|---|---|---|---|---|
| Accelerators | 528 accelerators | 2024-11 | Operator stated | Harvard Gazette: 528 GPUs in the cluster at the Nov 2024 Green500 ranking (H100s benchmarked). Pre-2026 size; a later expansion is not captured here. | news.harvard.edu |
The read
Harvard's Kempner Institute's academic AI training cluster at the MGHPCC in Holyoke, Massachusetts. It began as a spring 2023 pilot of 144 A100 (40 GB) GPUs and added 384 H100 (80 GB) GPUs in fall 2023, then in March 2026 announced an expansion to a stated 1,144 GPUs (424 H200, 384 H100, 144 A100 and 192 RTX PRO 6000 Blackwell) and a 1.8 exaflops BF16 peak. The Kempner handbook lists H100 and H200 nodes as 4 GPUs each and RTX6000 nodes as 8. The HPL figures (16.29 PFlop/s Rmax, 20.02 PFlop/s Rpeak) are the Lenovo ThinkSystem SD665-N V3 H100 partition submitted to TOP500 in November 2024 (rank 85, later 101, 114 and 133 on the 2025 and June 2026 lists) with AMD EPYC 9454 hosts and InfiniBand HDR; both TOP500 and the Harvard Gazette state the 16.29 PFlop/s figure and the rank of 85, so the tier is verified for those figures. The other partitions were not part of that HPL run. The Kempner AI cluster page and the MGHPCC project page give differing H200 counts (424 versus 384 or 40), so no H200 quantity is recorded.
Timeline
- 2023 Announced source
- 2023 First operational source
- 2024-11 List debut at #85 (November 2024) source
Public list appearances
Pointers only: rank and edition, linking to the canonical entry. We do not reproduce list tables. See sourcing policy.
Site & facility
Change history
- 2026-10-01 Added Kempner AI Cluster (Kempner Institute)
Source check
We fetched this system's own citations and recorded whether each page actually mentions it. This is a corroboration signal, not a fact check, and it is published so you can see how well the sourcing holds up rather than take it on trust.
| Cited source | Result | Found on the page |
|---|---|---|
| news.harvard.edu | names system, part and count | NVIDIA H100 SXM5, NVIDIA A100 SXM4 40GB · counts: 384, 144 |
| top500.org | names system and part | NVIDIA H100 SXM5, NVIDIA InfiniBand HDR |
Checked 2026-10-09 by pnpm verify. Re-run it and the table changes with the web.
Further reading
Operating a supercomputer
Acceptance testing, why "installed" and "in production" are six months apart, and the machine lifecycle.
Inside a node
Latency-optimised versus throughput-optimised processors, NUMA, and the end of the host-to-device copy.
The interconnect
Topologies, why latency and tail behaviour matter more than bandwidth, and how the fabric market consolidated.