Supercomputers/Systems/MN-3
Japan · operational · ai training
MN-3
Operated by Preferred Networks at Preferred Networks (Yokohama) .
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
- 2.18 PFlop/s
- Rpeak
- 3.35 PFlop/s
- Rmax ÷ Rpeak
- 65.1%
- Cores
- 1,664
- Power
- 53.29 kW
- Per watt
- 40.9 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 | - MN-Core Server, self-built by Preferred Networks | Preferred Networks | - | Reported | preferred.jptop500.org |
| Accelerator | Preferred Networks MN-Core (1st generation) Accelerators TSMC 12nm · 500 W Preferred Networks MN-Core, 128 processors used in the record-setting Green500 run | Preferred Networks | 128 reported | Verified | preferred.jptop500.org |
| CPU | Intel Xeon Platinum 8260M (Cascade Lake) CPUs x86-64 · 24C · 2.4 GHz Xeon Platinum 8260M 24C 2.4GHz | Intel | - | Reported | top500.orgprojects.preferred.jp |
| Interconnect | MN-Core DirectConnect Interconnect 112 Gb/s per link MN-Core DirectConnect | Preferred Networks | - | Reported | top500.orgprojects.preferred.jp |
The read
The only system in this dataset whose accelerator role is filled by neither NVIDIA nor AMD: MN-Core is a deep-learning ASIC that Preferred Networks co-developed with Kobe University, built around a matrix-arithmetic grid with no conventional cache hierarchy or branch prediction. MN-3 topped the Green500 list four times running, June 2020, November 2020, June 2021 and November 2021, reaching 39.38 GFlops/Watt on its last win, all in a window before this dataset's Green500 tracking otherwise begins in November 2023, which is why the rank pointer here targets the November 2021 edition specifically. Preferred Networks' own release and TOP500's independent Green500 listing agree on the rank, efficiency and power figures, so the tier is verified. Its current operational status has not been independently confirmed since its last public TOP500 appearance in mid 2024; we carry the last documented state rather than inferring a retirement.
Timeline
- 2020 Announced source
- 2020-05 First operational source
- 2021-11 List debut at #1 (November 2021) 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
Preferred Networks (Yokohama)
- Location
- Yokohama, Japan
- Commissioned
- 2020
Change history
- 2026-09-24 MN-3 added (Preferred Networks, Japan)
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 |
|---|---|---|
| top500.org | names system and part | Intel Xeon Platinum 8260M (Cascade Lake), Preferred Networks MN-Core (1st generation), MN-Core DirectConnect |
| preferred.jp | names system, part and count | Intel Xeon Platinum 8260M (Cascade Lake), Preferred Networks MN-Core (1st generation), MN-Core DirectConnect · counts: 128 |
Checked 2026-10-09 by pnpm verify. Re-run it and the table changes with the web.
Further reading
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.
Operating a supercomputer
Acceptance testing, why "installed" and "in production" are six months apart, and the machine lifecycle.