Parts/Accelerators/AMD Instinct MI300A
Why this part
An APU at exascale: three Zen 4 chiplets and six CDNA 3 chiplets in one package over one pool of 128GB HBM3, with no host memory, no device memory and no copy between them. El Capitan is built on it. Where GH200 links a CPU and a GPU coherently, MI300A stops treating them as separate things at all, and for codes that move irregularly between scalar and parallel phases that is a larger change than any arithmetic figure.
Systems using this part
- Systems
- 15
- Rmax underneath
- 2.81 EFlop/s
- Units recorded
- 2,118
8 of 15 edges disclose a quantity
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| HPC7 2026-06 | Accelerator | Italy | - | 571.5 PFlop/s | AMD | #6 → |
| Sirius (Tsukuba) 2026 | Accelerator | Japan | 96 | 5.63 PFlop/s | AMD | - |
| Plasma Simulator Sosei 2025-07 | Accelerator | Japan | 280 | - | AMD | - |
| bwUniCluster 3.0 2025-04 | Accelerator | Germany | 4 | - | AMD | - |
| Viper-GPU 2025-02 | Accelerator | Germany | 654 | 31.10 PFlop/s | AMD | #55 → |
| Tuolumne 2025-01 | Accelerator | USA | - | 208.1 PFlop/s | AMD | #14 → |
| Hunter 2025-01 | Accelerator | Germany | 752 | 31.68 PFlop/s | AMD | #54 → |
| Cosmos (SDSC) 2025 | Accelerator | USA | 168 | 5.18 PFlop/s | AMD | #272 → |
| El Capitan 2024-11 | Accelerator | USA | - | 1.81 EFlop/s | AMD | #2 → |
| El Dorado 2024-11 | Accelerator | USA | - | 68.02 PFlop/s | AMD | #41 → |
| rzAdams 2024 | Accelerator | USA | - | 24.38 PFlop/s | AMD | #87 → |
| Blueback 2024 | Accelerator | USA | 128 | 10.30 PFlop/s | AMD | #139 → |
| AMD Ouranos 2024 | Accelerator | France | - | 2.99 PFlop/s | AMD | #9 → |
| Portage 2024 | Accelerator | USA | - | 24.50 PFlop/s | AMD | #10 → |
| Santos Dumont | Accelerator | Brazil | 36 | 14.29 PFlop/s | AMD | #52 → |
“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.
Related news
- El Capitan verified as the fastest system in the world · Lawrence Livermore National LaboratoryLLNL confirms the MI300A-based system past 1.7 EFlop/s, making it the third US exascale machine and the second built entirely on AMD silicon.