Supercomputers/Systems/Niflheim
Denmark · operational · mixed
Niflheim
Also known as Niflheim Linux supercomputer cluster; DTU Physics Niflheim
Operated by Technical University of Denmark .
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: 1263 of 1345 systems have a readable citation that names them, and 56 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
- -
- Rpeak
- 2.38 PFlop/s
- Rmax ÷ Rpeak
- -
- Cores
- 31,160
- 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 | - Lenovo supplied the newest nodes (SD665 V3, SD650-N V3, SD650-N V2 and SR850 V3 servers) | Lenovo | - | Reported | niflheim-users.readthedocs.io |
| Integrator | - Dell supplied the Ice Lake, Cascade Lake and Skylake nodes and the MI210 node | Dell Technologies | - | Reported | niflheim-users.readthedocs.io |
| Accelerator | NVIDIA H200 SXM Accelerators CUDA (Hopper GH100) · 16,896C · 1.98 GHz · TSMC 4N · 700 W NVIDIA H200 SXM4 141 GB GPUs, four in each of two Lenovo SD650-N V3 nodes | NVIDIA | 8 reported | Reported | niflheim-users.readthedocs.io |
| Accelerator | NVIDIA A100 SXM4 40GB Accelerators CUDA (Ampere GA100) · 6,912C · 1.41 GHz · 7 nm · 400 W NVIDIA A100 SXM4 40 GB GPUs, four in each of four Lenovo SD650-N V2 nodes | NVIDIA | 16 reported | Reported | niflheim-users.readthedocs.io |
| Accelerator | AMD Instinct MI210 Accelerators ROCm (CDNA 2) · TSMC N6 · 300 W AMD Instinct MI210 64 GB GPUs in one Dell R7525 node | AMD | 2 reported | Reported | niflheim-users.readthedocs.io |
| CPU | AMD EPYC 9004 (Genoa) CPUs x86-64 · 96C · 2.4 GHz · TSMC N5 · 360 W AMD EPYC 9474F 48-core 3.6 GHz, two per node in 80 Lenovo SD665 V3 nodes | AMD | 160 reported | Reported | niflheim-users.readthedocs.io |
| CPU | - Intel Xeon Gold 6348 28-core Ice Lake, two per node in 96 Dell R650 nodes, plus older Cascade Lake and Skylake nodes | Intel | - | Reported | niflheim-users.readthedocs.io |
| Interconnect | NVIDIA InfiniBand NDR (Quantum-2) Interconnect Fat tree or dragonfly+ · 400 Gb/s per port · Under 0.6 microseconds NVIDIA ConnectX-7 NDR200 InfiniBand on the EPYC 9474F nodes | NVIDIA | - | Reported | niflheim-users.readthedocs.io |
| Interconnect | Omni-Path Architecture Interconnect Fat tree · 100 Gb/s per port · About 1 microsecond Cornelis Networks Omni-Path 100 Gbit/s on the Ice Lake nodes | Cornelis Networks | - | Reported | niflheim-users.readthedocs.io |
The read
The Linux cluster of the Department of Physics at the Technical University of Denmark, available to DTU staff and students and named in a 2002 naming competition at the CAMP research centre. Its user documentation says that as of March 2026 it has 698 compute nodes with 31,160 CPU cores and 106 GPUs, more than 2,380 TFlop/s aggregate theoretical CPU peak and 313 TB of RAM, built from several generations: 80 Lenovo SD665 V3 nodes with two AMD EPYC 9474F 48-core CPUs on NDR200 InfiniBand (installed February 2024 to December 2025), two Lenovo nodes with four NVIDIA H200 GPUs each (July 2025), four Lenovo 4 TB nodes, four Lenovo nodes with four NVIDIA A100 40GB GPUs each, a Dell node with two AMD Instinct MI210 GPUs, 96 Dell R650 Ice Lake nodes on Omni-Path (December 2021) and older Dell Cascade Lake and Skylake, SuperMicro GPU and Huawei Broadwell nodes, under Rocky Linux with Slurm. The documentation is the only source and gives the CPU peak, not an Rmax, so the tier is reported and no TOP500 appearance is recorded.
Timeline
No dated events recorded.
Change history
- 2026-10-09 Niflheim added from its source pages
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 |
|---|---|---|
| niflheim-users.readthedocs.io | names system, part and count | AMD EPYC 9004 (Genoa), NVIDIA H200 SXM, NVIDIA A100 SXM4 40GB, AMD Instinct MI210, NVIDIA InfiniBand NDR (Quantum-2) · counts: 160, 8, 16, 2 |
Checked 2026-10-13 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.