Supercomputers/Systems/Berzelius
Sweden · operational · ai training
Berzelius
Operated by NSC at NSC Data Centre, Linkoping .
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
- 5.25 PFlop/s
- Rpeak
- 9.64 PFlop/s
- Rmax ÷ Rpeak
- 54.5%
- Cores
- 59,520
- 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 | - Bull-built NVIDIA DGX SuperPOD (per TOP500 manufacturer field) | Bull | - | Reported | top500.org |
| Accelerator | NVIDIA A100 SXM4 40GB Accelerators CUDA (Ampere GA100) · 6,912C · 1.41 GHz · 7 nm · 400 W 60 thin DGX A100 nodes with 8 A100 40 GB GPUs each | NVIDIA | 480 reported | Reported | nsc.liu.se |
| Accelerator | - 34 fat DGX A100 nodes with 8 A100 80 GB GPUs each | NVIDIA | 272 reported | Reported | nsc.liu.se |
| Accelerator | NVIDIA H200 SXM Accelerators CUDA (Hopper GH100) · 16,896C · 1.98 GHz · TSMC 4N · 700 W Berzelius Hopper: 16 DGX H200 nodes with 8 H200 141 GB GPUs each | NVIDIA | 128 reported | Reported | nsc.liu.se |
| CPU | AMD EPYC 7742 (Rome) CPUs x86-64 · 64C · 2.25 GHz · 7 nm · 225 W Berzelius Ampere: 94 DGX A100 nodes, two AMD EPYC 7742 each | AMD | - | Reported | nsc.liu.setop500.org |
| CPU | Intel Xeon Platinum 8480C CPUs x86-64 · 56C · 2 GHz · Intel 7 · 350 W Berzelius Hopper DGX H200 nodes: two Intel Xeon 8480C each | Intel | - | Reported | nsc.liu.se |
| Interconnect | NVIDIA InfiniBand HDR Interconnect Fat tree or dragonfly+ · 200 Gb/s per port · About 0.6 microseconds 8x NVIDIA Mellanox HDR per DGX A100, non-blocking fat-tree | NVIDIA | - | Reported | nsc.liu.setop500.org |
| Interconnect | NVIDIA InfiniBand NDR (Quantum-2) Interconnect Fat tree or dragonfly+ · 400 Gb/s per port · Under 0.6 microseconds 8x NVIDIA Mellanox NDR per DGX H200, non-blocking fat-tree | NVIDIA | - | Reported | nsc.liu.se |
| Storage | VAST Data Platform Storage Disaggregated shared-everything 3 PB VAST Data cluster with NVMe-oF | VAST Data | - | Reported | nsc.liu.se |
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 | 880 accelerators | 2026-10 | Operator stated | 752 A100 (94 DGX-A100 nodes) plus 128 H200 (16 DGX-H200 nodes) per NSC; sum is our addition. | nsc.liu.se |
The read
Sweden's AI-focused NVIDIA DGX SuperPOD at the National Supercomputer Centre in Linkoping, installed in spring 2021 with 60 DGX A100 nodes, expanded in 2023 to 94 DGX A100 nodes (60 with 40 GB and 34 with 80 GB A100 GPUs, HDR InfiniBand) and again with Berzelius Hopper, 16 DGX H200 nodes on NDR InfiniBand, plus a 3 PB VAST Data storage cluster. TOP500 lists the system as built by Bull and ranked it 82nd on the June 2021 list at 5.25 PFlop/s Rmax on 59,520 cores; those TOP500 figures describe the 2021 configuration and are not restated for the expanded machine.
Timeline
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
- 2026-10-04 Added AI datacenter metrics for berzelius
- 2026-10-01 Added Berzelius (NSC)
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 | AMD EPYC 7742 (Rome), NVIDIA A100 SXM4 40GB, NVIDIA InfiniBand HDR, NVIDIA InfiniBand NDR (Quantum-2) |
| nsc.liu.se | names system, part and count | AMD EPYC 7742 (Rome), NVIDIA A100 SXM4 40GB, NVIDIA H200 SXM, VAST Data Platform · counts: 480 |
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.