Parts/CPUs/AMD EPYC 7763 (Milan)
Why this part
Milan kept Rome's chiplet packaging and reorganised the cache so that all eight cores in a complex share one 32MB L3 instead of splitting into two pools of four. That halves the cases where two cooperating threads land on opposite sides of a cache boundary, which matters far more to real scientific code than the modest clock increase. Perlmutter pairs it with A100s at NERSC.
Systems using this part
- Systems
- 26
- Rmax underneath
- 184.8 PFlop/s
- Units recorded
- 1,925
7 of 26 edges disclose a quantity
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| NASA Cabeus 2023-12 | CPU | USA | 128 | - | AMD | - |
| Derecho 2023-07 | CPU | USA | - | 10.32 PFlop/s | AMD | #59 → |
| Supek 2023 | CPU | Croatia | - | - | AMD | - |
| Dhabi Supercomputer 2023 | CPU | United Arab Emirates | - | 5.09 PFlop/s | AMD | #174 → |
| Komondor 2022-12 | CPU | Hungary | - | 3.10 PFlop/s | AMD | - |
| Delta 2022-11 | CPU | USA | 132 | 3.81 PFlop/s | AMD | #141 → |
| Zaratan 2022-08 | CPU | USA | - | - | AMD | - |
| SK Telecom Titan 2022-05 | CPU | South Korea | - | 19.53 PFlop/s | AMD | #92 → |
| Champollion 2022-05 | CPU | France | - | 2.32 PFlop/s | AMD | #10 → |
| Anvil 2022-02 | CPU | USA | - | 3.56 PFlop/s | AMD | #173 → |
| Noctua 2 2022 | CPU | Germany | - | 4.19 PFlop/s | AMD | #121 → |
| Topaze 2021-06 | CPU | France | 864 | 3.26 PFlop/s | AMD | #201 → |
| Perlmutter 2021-05 | CPU | USA | - | 79.23 PFlop/s | AMD | #28 → |
| Levante 2021 | CPU | Germany | - | 10.11 PFlop/s | AMD | #53 → |
| COSMA8 2021 | CPU | United Kingdom | 168 | - | AMD | - |
| CEA-HF 2021 | CPU | France | - | 23.24 PFlop/s | AMD | #14 → |
| Wilkes-3 2021 | CPU | United Kingdom | - | 2.29 PFlop/s | AMD | #3 → |
| Hortense 2021 | CPU | Belgium | 384 | - | AMD | - |
| Aitken 2019 | CPU | USA | - | 9.07 PFlop/s | AMD | #58 → |
| Kebnekaise 2016-11 | CPU | Sweden | 1 | - | AMD | - |
| Euler 2014 | CPU | Switzerland | 248 | - | AMD | - |
| Rocket Cluster | CPU | Estonia | - | - | AMD | - |
| Agate | CPU | USA | - | 1.88 PFlop/s | AMD | #379 → |
| Lonestar6 | CPU | USA | - | - | AMD | - |
| hotlum | CPU | France | - | 3.81 PFlop/s | AMD | #142 → |
| Hábrók | CPU | Netherlands | - | - | AMD | - |
“Units recorded” sums only the edges where a public source states a quantity. It is a floor, not a total, and should never be read as installed base.