# Frontier: every supplier in the machine

Oak Ridge National Laboratory · US · system record on Compute Atlas: https://computeatlas.net/supercomputers/systems/frontier

## Specs

- **accelerator:** AMD Instinct MI250X, 4 per node (2 GCDs per module, 8 GCDs per node), 128 GB HBM2E per module; TSMC 6nm
- **host cpu:** 1x 64-core AMD Optimized 3rd Gen EPYC per node, 2.0 GHz base, 3.7 GHz boost
- **integrator:** HPE (HPE Cray EX supercomputer; TOP500 lists the system as HPE Cray EX235a)
- **nodes:** 9,856 compute nodes in 77 cabinets of 128 nodes each (OLCF user guide)
- **memory:** Per node: 512 GB DDR4 plus 4 x 128 GB HBM2E (8 GCDs x 64 GB)
- **interconnect:** HPE Slingshot-11, 4 x 200 Gb/s NICs per node (100 GB/s injection), 64-port switches, dragonfly topology
- **storage:** Orion: 679 PB usable Lustre namespace on HPE Cray ClusterStor E1000 (three tiers: 11.5 PB NVMe performance, 679 PB HDD capacity, 10 PB NVMe metadata); 2 x 1.92 TB NVMe per compute node
- **cooling:** Direct liquid cooling in the cabinets via coolant distribution units; facility warm-water plant with four evaporative cooling towers, 90°F design supply water, sized for 40 MW
- **power mw:** 24.6 MW (24,607 kW, power measurement level 2, TOP500 June 2026 list)
- **rmax pflops:** 1,353.00 PFlop/s (TOP500 June 2026 list)
- **rpeak pflops:** 2,055.72 PFlop/s (TOP500 June 2026 list)
- **status note:** First listed on the June 2022 TOP500 list (ORNL announcement, 2022-05-30); TOP500 gives the installation year as 2021; operated by the OLCF for DOE's Office of Science

Unmarked companies come from the operator's own pages (ornl.gov, olcf.ornl.gov, docs.olcf.ornl.gov), ORNL-authored papers or the vendor's own materials (hpe.com-hosted briefings, amd.com); 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: ORNL/OLCF documentation (docs.olcf.ornl.gov, ornl.gov news), an HPE architecture briefing hosted by OLCF, AMD's MI250X pages, ORNL's CUG 2023 Orion paper, an ORNL-authored Scientific Data paper on Frontier's cooling, and the TOP500 entry (June 2026 list), researched 2026-10-02

## Lines

1. System design and cabinets (HPE Cray EX, liquid-cooled, up to 64 compute blades in 8 chassis per cabinet)
   - parts: Rack-scale systems (4.01)
   - companies: HPE (system integrator and cabinet designer (HPE Cray EX))
   - sources: https://docs.olcf.ornl.gov/systems/frontier_user_guide.html https://www.olcf.ornl.gov/wp-content/uploads/2-15-23-Frontier-System-Architecture-public-v7.pdf
2. Compute blade (HPE Cray EX235A: two nodes per blade, each with 1 CPU and 4 MI250X, all components liquid-cooled)
   - parts: GPU baseboards & compute trays (4.03)
   - companies: HPE
   - not confirmed: maker of the node's DDR4 DIMMs (512 GB per node) — not named in any source read; maker of the two node-local NVMe drives (1.92 TB each) — not named in any source read
   - sources: https://www.olcf.ornl.gov/wp-content/uploads/2-15-23-Frontier-System-Architecture-public-v7.pdf
3. MI250X accelerator (two GCDs per module, fabricated on TSMC 6nm; 4 modules per node)
   - parts: Merchant GPUs / accelerators (1.01); Leading-edge foundry (1.09)
   - companies: AMD; TSMC (6nm FinFET process, per AMD's MI250X specification page)
   - not confirmed: maker of the eight HBM2E stacks per module (128 GB) — AMD and ORNL give capacity and bandwidth but no memory supplier
   - sources: https://www.amd.com/en/products/accelerators/instinct/mi200/mi250x.html https://www.amd.com/en/newsroom/press-releases/2021-11-8-new-amd-instinct-mi200-series-accelerators-bring-.html
4. Two-die package with 2.5D Elevated Fanout Bridge (EFB) interconnect and eight HBM2E stacks
   - parts: 2.5D/3D advanced packaging (1.11); HBM (2.01)
   - companies: AMD (announced the EFB packaging technology with the MI200 series); TSMC* (described as helping AMD develop the elevated fanout bridge)
   - not confirmed: which company performs the final package assembly and test; HBM2E supplier (see MI250X line)
   - sources: https://www.amd.com/en/newsroom/press-releases/2021-11-8-new-amd-instinct-mi200-series-accelerators-bring-.html https://www.nextplatform.com/2021/11/09/the-aldebaran-amd-gpu-that-won-exascale/
5. Host CPU (one 64-core AMD Optimized 3rd Gen EPYC per node, coherent with the GPUs over Infinity Fabric)
   - parts: Server CPUs (1.05)
   - companies: AMD
   - not confirmed: foundry for the custom EPYC part — no source read names it
   - sources: https://docs.olcf.ornl.gov/systems/frontier_user_guide.html
6. Network interface (four HPE Slingshot 200 Gb/s NICs per node, 100 GB/s injection bandwidth)
   - parts: NICs / SuperNICs (1.06)
   - companies: HPE (Slingshot NICs)
   - sources: https://docs.olcf.ornl.gov/systems/frontier_user_guide.html
7. Slingshot switches (64-port Rosetta ASIC on TSMC 16nm; 74 groups of 32 switches in a dragonfly)
   - parts: Scale-out switch ASICs (3.02); Switch systems & white-box ODMs (3.03); Leading-edge foundry (1.09)
   - companies: HPE (designs and supplies the Slingshot switches); TSMC (Rosetta switch ASIC process, per a paper co-authored by HPE engineers)
   - not confirmed: maker of the copper and active optical cables (copper inside groups, optics between groups) — the cable manufacturer is not named in any source read
   - sources: https://www.olcf.ornl.gov/wp-content/uploads/2-15-23-Frontier-System-Architecture-public-v7.pdf https://arxiv.org/abs/2008.08886
8. Orion parallel file system (Lustre on HPE Cray ClusterStor E1000; 679 PB usable)
   - parts: AI storage systems (2.10)
   - companies: HPE (ClusterStor E1000 hardware and integration of Orion)
   - sources: https://www.ornl.gov/news/frontier-supercomputer-debuts-worlds-fastest-breaking-exascale-barrier https://www.olcf.ornl.gov/wp-content/uploads/2-15-23-Frontier-System-Architecture-public-v7.pdf
9. Orion drives (Seagate Exos X18 18 TB HDDs, Samsung PM1733 NVMe performance tier, KIOXIA 30 TB NVMe metadata tier)
   - parts: HDD (2.09); Drives (2.08a)
   - companies: Seagate (18 TB Exos X18 hard drives, 106 per enclosure, two enclosures per scalable storage unit); Samsung Electronics (3.2 TB PM1733 NVMe SSDs in each storage unit (performance tier)); Kioxia (30 TB NVMe SSDs in the metadata units)
   - sources: https://cug.org/proceedings/cug2023_proceedings/includes/files/pap122s2-file1.pdf
10. In-cabinet liquid cooling (direct cold plates on all node components, shared coolant distribution units; 25 CDUs, each serving three cabinets)
   - parts: Direct-to-chip cold plates (5.01); Coolant distribution units (5.05)
   - companies: HPE (cabinet, blade cold plates and CDUs are part of the HPE Cray EX design)
   - not confirmed: makers of the cold plates, quick disconnects and CDU internals — not named in any source read; Orion's rear-door heat exchangers are mentioned in ORNL's CUG 2023 paper but no manufacturer is named
   - sources: https://www.olcf.ornl.gov/wp-content/uploads/2-15-23-Frontier-System-Architecture-public-v7.pdf https://www.nature.com/articles/s41597-024-03913-w
11. Cabinet power (four power shelves per cabinet, up to eight AC/DC rectifiers each)
   - parts: Power shelves & PSUs (4.12)
   - companies: HPE
   - not confirmed: rectifier manufacturer — not named in any source read
   - sources: https://www.olcf.ornl.gov/wp-content/uploads/2-15-23-Frontier-System-Architecture-public-v7.pdf
12. Facility cooling plant and its building automation system (four evaporative cooling towers, four 350 hp pumps, 40 MW design load)
   - parts: Building management, controls, environmental sensors (6.08)
   - companies: Johnson Controls (Metasys building automation system that logs Frontier's flow, temperature and power)
   - not confirmed: manufacturers of the cooling towers, pumps and plate heat exchangers — ORNL describes them without naming vendors; manufacturers of the nine 4 MW transformers and the switchgear in the computer room (Layer 7) — not named
   - sources: https://www.nature.com/articles/s41597-024-03913-w https://www.olcf.ornl.gov/2021/09/29/stunning-specs-whats-inside-the-nations-first-exascale-supercomputer-facility/

* reported: named trade press or analyst reporting rather than the operator or vendor.

## What could not be confirmed

HPE and ORNL do not itemise the compute side's own components — no source read names the maker of the HBM2E stacks, the DDR4 DIMMs, the node-local NVMe drives, the cabinet cold plates, the rectifiers or the Slingshot cables and optics. The facility's 40 MW electrical feed, nine 4 MW transformers, four cooling towers and 350-horsepower pumps are described by ORNL without naming their manufacturers, so Layers 6 and 7 are only partly mapped here (the one named facility vendor is the building automation system). The cabinet and node counts also differ between ORNL documents, as noted in the specs.