130 parts · 1,202 companies
Prices as of2026-10-25
Search

LUMI: every supplier in the machine

CSC IT Center for Science (Kajaani), Finland

Accelerator
AMD Instinct MI250X, 4 per LUMI-G node (11,912 GPUs in 2,978 nodes), 128 GB HBM2e per module; TSMC 6nm
Host CPU
LUMI-G: 1 x 64-core AMD EPYC 7A53 "Trento" per node; LUMI-C: 2 x AMD EPYC 7763 "Milan" (64 cores each) per node
Integrator
HPE (HPE Cray EX supercomputer; TOP500 lists the system as HPE Cray EX235a)
Nodes
2,978 LUMI-G GPU nodes, 2,048 LUMI-C CPU nodes and 16 LUMI-D large-memory nodes
Memory
LUMI-G node: 512 GiB DDR4 plus 4 x 128 GB HBM2e; LUMI-C node: 256, 512 or 1,024 GiB
Interconnect
HPE Slingshot-11, 200 Gb/s NICs (4 per LUMI-G node, 1 per LUMI-C node), dragonfly topology
Storage
About 117 PB: LUMI-P 4 x 20 PB Lustre on hard disks, LUMI-F 8 PB all-flash Lustre, LUMI-O 30 PB Ceph object storage
Cooling
Direct liquid cooling in the HPE Cray EX cabinets; waste heat reused in Kajaani district heating through heat pumps, with free cooling for much of the year
Power
7.1 MW (7,106.82 kW, power measurement level 2, TOP500 June 2026 list)
Rmax (HPL)
379.70 PFlop/s (TOP500 June 2026 list)
Rpeak
531.51 PFlop/s (TOP500 June 2026 list)
Status
First listed in June 2022 as a partial system (151.9 PFlop/s); the full 379.70 PFlop/s configuration appears from the November 2023 list; TOP500 gives the installation year as 2023; run by CSC for the LUMI consortium and the EuroHPC JU

Unmarked companies come from the operator's or a vendor's own pages (lumi-supercomputer.eu, docs.lumi-supercomputer.eu, csc.fi, eurohpc-ju.europa.eu, amd.com, hpe.com, top500.org); reported companies come from named trade press whose article body was read in full, not just the headline. This is a map of who plays where, not a confirmed bill of materials: LUMI documentation (docs.lumi-supercomputer.eu), CSC and LUMI news pages, the EuroHPC JU supercomputer page, a CSC heat-reuse presentation, an HPE-hosted storage article, AMD's MI250X page and the TOP500 system entry (June 2026 list), plus one IT Pro article, researched 2026-10-03

This page is the parts-first view. For LUMI's full system record on the sister site Compute Atlas, an open dataset of supercomputer and AI-cluster supply chains, see LUMI on Compute Atlas.

  1. System design and cabinets (HPE Cray EX liquid-cooled cabinets; eight compute chassis of up to eight blades each)
    Rack-scale systems 4.01
    HPE (system integrator and cabinet designer (HPE Cray EX))
    Sources: docs.lumi-supercomputer.eu · assets.ext.hpe.com · itpro.com
  2. LUMI-G compute blade (HPE Cray EX235a node type; one 64-core EPYC Trento and four MI250X per node; 2,978 nodes)
    GPU baseboards & compute trays 4.03
    HPE (Cray EX blade and node design) · maker of the node's DDR4 DIMMs (512 GiB per node) — not named in any source read, maker of the blade board and who assembles it — not named in any source read
    Sources: top500.org · docs.lumi-supercomputer.eu
  3. AMD Instinct MI250X accelerator (two GCDs per module, 110 compute units each, 64 GB HBM2e per GCD; TSMC 6nm)
    Merchant GPUs / accelerators 1.01 · Leading-edge foundry 1.09
    AMD (designer; chosen for LUMI on performance per watt, per the LUMI director), TSMC (6nm FinFET process, per AMD's MI250X specification page) · maker of the HBM2e stacks (128 GB per module) — AMD and LUMI give capacity and bandwidth but name no memory supplier
    Sources: amd.com · docs.lumi-supercomputer.eu · lumi-supercomputer.eu
  4. LUMI-G host CPU (AMD EPYC 7A53 "Trento", 64 Zen 3 cores, wired to the GPUs by Infinity Fabric)
    Server CPUs 1.05
    AMD · foundry for the custom EPYC 7A53 part — no source read names it
    Sources: docs.lumi-supercomputer.eu
  5. LUMI-C CPU partition (2,048 nodes, each with two 64-core AMD EPYC 7763 "Milan"; 256 GiB to 1 TiB per node)
    Server CPUs 1.05
    AMD · maker of the node DDR4 memory — not named in any source read
    Sources: docs.lumi-supercomputer.eu · lumi-supercomputer.eu
  6. LUMI-D visualization nodes (eight nodes with 2 x AMD EPYC 7742 and 8 x NVIDIA A40 each; 64 A40 GPUs in total)
    Merchant GPUs / accelerators 1.01 · Server CPUs 1.05
    Nvidia (A40 visualization GPUs; the docs say their main purpose is visualization, not GPU computing), AMD (EPYC 7742 (Zen 2 "Rome") host CPUs)
    Sources: docs.lumi-supercomputer.eu · lumi-supercomputer.eu
  7. Network interface (200 Gb/s HPE Slingshot NIC; four per LUMI-G node, one per MI250X module, 800 Gb/s injection)
    NICs / SuperNICs 1.06
    HPE (Slingshot NICs)
    Sources: docs.lumi-supercomputer.eu · lumi-supercomputer.eu
  8. Slingshot switches and dragonfly fabric (LUMI-G: 24 groups of 32 switches; LUMI-C: 8 groups of 16 switches; copper inside groups, optical between)
    Scale-out switch ASICs 3.02 · Switch systems & white-box ODMs 3.03 · Leading-edge foundry 1.09
    HPE (designs and supplies the Slingshot switches), TSMC (Rosetta switch ASIC process, per a paper co-authored by HPE engineers) · maker of the copper and optical cables and transceivers — the docs describe copper cables inside groups and optical cables between groups but name no cable maker
    Sources: docs.lumi-supercomputer.eu · arxiv.org
  9. LUMI-P main storage (four 20 PB Lustre file systems on hard disks, 240 GB/s each, on HPE Cray ClusterStor E1000)
    AI storage systems 2.10
    HPE (ClusterStor E1000 hardware) · maker of the hard drives — not named in any source read
    Sources: docs.lumi-supercomputer.eu · lumi-supercomputer.eu
  10. LUMI-F flash storage (8 PB all-flash Lustre with 58 OSTs and 2 metadata servers, 1,740 GB/s aggregate)
    AI storage systems 2.10
    HPE (ClusterStor E1000 all-flash system) · maker of the SSDs and the NVMe controllers — not named in any source read
    Sources: docs.lumi-supercomputer.eu · lumi-supercomputer.eu
  11. LUMI-O object storage (30 PB of Ceph on HPE Apollo 4200 Gen10 servers, S3 access, separate from the compute fabric)
    AI storage systems 2.10
    HPE (Apollo 4200 Gen10 servers, per HPE's own 2021 storage article written before installation) · maker of the drives in the Apollo servers — not named in any source read
    Sources: community.hpe.com · docs.lumi-supercomputer.eu
  12. In-cabinet direct liquid cooling (cabinet cooling manifolds, quick-connect blades and a coolant distribution unit for the Cray EX cabinets)
    Direct-to-chip cold plates 5.01 · Coolant distribution units 5.05
    HPE (cabinet, blade cold plates and CDUs are part of the HPE Cray EX design) · makers of the cold plates, quick disconnects and CDU internals — not named in any source read, LUMI's own CDU count and secondary-loop temperatures — not found; the HPE brochure figures are generic Cray EX values
    Sources: assets.ext.hpe.com · kamk.fi · itpro.com
  13. Hydroelectric electricity supply (100% hydropower, supplied by Vattenfall according to CSC's 2021 data-centre article)
    Utilities, IPPs & power providers signing data-center deals 8.07
    Vattenfall (named as supplier in a 2021 pre-opening CSC article ("will be covered"); the current supply contract was not checked)
    Sources: lumi-supercomputer.eu

No source read names the maker of the HBM2e stacks, the DDR4 DIMMs, the SSDs in LUMI-F, the hard drives in LUMI-P, the Slingshot cables and optics, or the cold plates, quick disconnects, rectifiers and coolant distribution units inside the cabinets. HPE's cabinet brochure is generic Cray EX material and says nothing LUMI-specific about power shelves, so no power line is listed. The Kajaani facility is described by CSC without naming equipment makers - three heat pumps (2,400 kW cooling and 3,000 kW heating each), 4,000 kW district-heating heat exchangers, dry-air backup cooling of about 10 MW and a computer hall in a former paper mill - so those are given here rather than as lines. The one named facility party is the district heating operator Loiste Lämpö, which is not in the catalog. LUMI's own page, EuroHPC and early HPE material give different flash-tier sizes and different node counts, as noted in the specs. Power and cooling outside the cabinet sit in Layer 6 and Layer 7.

Machine-readable: Markdown ·JSON ·every line of every machine (CSV)

Other machines:El Capitan · Frontier · Aurora · JUPITER · Alps · Leonardo · Isambard-AI · Venado · Perlmutter · Eagle (Microsoft Azure) · Eos · Fugaku · MareNostrum 5 · Horizon · Doudna (NERSC-10). See also the AI rack teardowns and thesupercomputer overview.