Supercomputers/Systems/Adastra 2
France · operational · mixed
Adastra 2
Also known as Adastra2; Adastra MI300A partition
Researched with AI assistance, checked by scripts. These rows were researched by AI agents reading public operator, laboratory and vendor sources, appended by scripts and checked by automated validation; no subject-matter expert has reviewed them (how this site is made). A mechanical second-reader pass has since fetched every cited source: 1350 of 1581 systems have a readable citation that names them, and 64 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
- 2.53 PFlop/s
- Rpeak
- 4.06 PFlop/s
- Rmax ÷ Rpeak
- 62.3%
- Cores
- 16,128
- Power
- 36.6 kW
- Per watt
- 69.1 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.
How it compares
Ranked against the other machines in this dataset that report the same figure: those that first came into service within two years of Adastra 2, and those in France. Computed from our tables; a metric is ranked only when at least eight peers report it.
| Metric | Value | Against its era | Against its country |
|---|---|---|---|
| Rmax | 2.53 PFlop/s | #222 of 234 · bottom 6% | #22 of 52 · top 41% |
| Energy efficiency | 69.1 GF/W | #1 of 121 · highest | #1 of 42 · highest |
| Rmax ÷ Rpeak | 62.3 % | #189 of 231 · bottom 19% | #44 of 51 · bottom 15% |
| Power draw | 37 kW | #125 of 126 · bottom 2% | #42 of 42 · bottom 1% |
Every metric is ranked largest-first, so a high power-draw rank means a big power draw, not a good one. Rmax ÷ Rpeak uses HPL-measured results only. Rankings move as machines are added; they describe this dataset, not the world.
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 | - HPE Cray EX4000 cabinet with 14 EX255a blades | Hewlett Packard Enterprise | - | Verified | hpcwire.comtop500.org |
| Accelerator | AMD Instinct MI300A Accelerators ROCm (CDNA 3) · TSMC N5/N6 · 550 W 28 nodes with 4 AMD Instinct MI300A APUs each | AMD | 112 reported | Verified | dci.dci-gitlab.cines.frhpcwire.com |
| CPU | AMD 4th Gen EPYC 24C (in MI300A) CPUs x86-64 · 24C · 1.8 GHz · TSMC N5 24 Zen 4 cores per MI300A APU (AMD 4th Gen EPYC 24C 1.8GHz) | AMD | - | Reported | top500.org |
| Interconnect | HPE Slingshot-11 Interconnect Dragonfly · 200 Gb/s per port · Around 2 microseconds HPE Slingshot-11, one 200 Gb/s NIC per APU | Hewlett Packard Enterprise | - | Verified | dci.dci-gitlab.cines.frhpcwire.com |
| Operating system | Red Hat Enterprise Linux Operating systems Linux distribution · Open source, commercial support subscription Red Hat Enterprise Linux | Red Hat | - | Reported | top500.org |
The read
The AMD Instinct MI300A extension of GENCI's Adastra, hosted and run by CINES in Montpellier; the MI250X/Genoa original is the adastra row. GENCI's announcement of 14 November 2023 (reproduced by HPCwire) planned 112 MI300A APUs in one new HPE Cray EX4000 cabinet holding 14 EX255a blades, i.e. 28 nodes with four APUs each and one 200 Gbps Slingshot NIC per APU, after GENCI and CINES decided on the extension in October 2023. CINES's architecture page now lists it as the Accelerated (MI300A) partition: 28 nodes, 512 GiB of HBM3 shared by CPU and GPU per node and 6.98 PFlop/s theoretical FP64 (the same page says HBM2 once, an internal inconsistency); it does not use the name Adastra 2, whereas TOP500 does. TOP500 (system 180319) lists 16,128 cores, 2.53 PFlop/s Rmax, 4.06 PFlop/s Rpeak and 36.6 kW at measurement level 1, installation year 2024, and ranks it 484th in June 2025; it is not on the November 2025 list. The TOP500 Rpeak (4.06) and CINES's theoretical FP64 (6.98) differ and neither source explains why. No exact go-live date was found, so only the installation year is recorded.
Timeline
- 2023-11 Announced source
- 2024 First operational source
- 2025-06 List debut at #484 (June 2025) source
Public list appearances
Pointers only: rank and edition, linking to the canonical entry. We do not reproduce list tables. See sourcing policy.
Change history
- 2026-10-29 Added Adastra 2
Nearby in scale
The machines closest to Adastra 2 by Rmax, smaller to larger.
- Rattler · USA, operational · 2.53 PFlop/s
- HEMUS · Bulgaria, operational · 2.53 PFlop/s
- Improv · USA, operational · 2.51 PFlop/s
- Cascade (PNNL) · USA, operational · 2.54 PFlop/s
- CMA PAI-B · China, operational · 2.55 PFlop/s
- JMA PRIMERGY CX2550 M7 · Japan, operational · 2.55 PFlop/s
Related systems
Same site
- Adastra · France, operational, 2022
- Occigen2 · France, decommissioned, 2017
- Occigen · France, upgraded, 2014
- Jade (CINES) · France, upgraded, 2008
Same operator
- Jean Zay H100 · France, operational, 2024
- Jean Zay V100 · France, operational, 2019
- Turing (IDRIS) · France, decommissioned, 2012
Same accelerator
- AMD Ouranos · France, operational, 2024
- Blueback · USA, operational, 2024
- El Capitan · USA, operational, 2024
- El Dorado · USA, operational, 2024
- Portage · USA, operational, 2024
- rzAdams · USA, operational, 2024
Further reading
Operating a supercomputer
Acceptance testing, why "installed" and "in production" are six months apart, and the machine lifecycle.
Inside a node
Why accelerators exist, what the memory hierarchy costs, and how the CPU and accelerator merged onto one package.
The interconnect
Topologies, why latency and tail behaviour matter more than bandwidth, and how the fabric market consolidated.