{
 "id": "alps",
 "url": "/supercomputers/alps",
 "name": "Alps",
 "site": "CSCS (Swiss National Supercomputing Centre)",
 "country": "CH",
 "compute_atlas_url": "https://computeatlas.net/supercomputers/systems/alps",
 "sourcing": "CSCS's Alps page and docs.cscs.ch user documentation, CSCS-authored papers and news, the 2021 CSCS/HPE/NVIDIA press release, HPE, NVIDIA, AMD and VAST Data pages, and the TOP500 entry (June 2026 list), researched 2026-10-03",
 "reading_note": "Unmarked companies come from the operator's or a vendor's own pages (cscs.ch, docs.cscs.ch, hpe.com, nvidia.com, amd.com, vastdata.com); reported companies come from named trade press or analyst reporting whose article body was read in full, not just the headline.",
 "specs": {
  "accelerator": "NVIDIA GH200 Grace Hopper Superchip (H100-class Hopper GPU, 96 GB HBM3 per GPU), 4 per node, 10,752 in total; other partitions use NVIDIA A100, AMD Instinct MI250X and AMD Instinct MI300A",
  "host_cpu": "GH200 partition: NVIDIA Grace, 72 Arm cores per superchip (TOP500: 3.1 GHz). Other partitions: AMD EPYC 7742 (Rome) in the CPU nodes, AMD EPYC hosts in the A100 and MI250X nodes, 24 Zen 4 cores per MI300A APU",
  "integrator": "HPE (HPE Cray EX; TOP500 lists the system as HPE Cray EX254n)",
  "nodes": "GH200 partition (headline): 2,688 nodes in 1,344 blades across 24 cabinets, 112 nodes per cabinet. Other partitions: 1,024 AMD Rome (EPYC 7742) nodes, 144 NVIDIA A100 nodes, 128 AMD MI300A nodes, 24 AMD MI250X nodes",
  "memory": "GH200 node: 4 x (128 GB LPDDR5X + 96 GB HBM3). Rome nodes: 256 GB or 512 GB DDR4. MI300A node: 512 GB unified HBM. Memory per A100 and MI250X host node differs between CSCS pages (see lines)",
  "interconnect": "HPE Slingshot-11, 200 Gb/s injection bandwidth per module/GPU; 4 NICs per GH200 node (one per Grace Hopper module), 2 NICs per Rome node",
  "storage": "Capstor: HPE ClusterStor E1000D Lustre, 129 PB raw (8,480 x 16 TB HDD). Iopsstor: ClusterStor E1000F Lustre, 7.2 PB raw (240 x 30 TB NVMe). Ritom: VAST, 10 PB (7.2 PB real), scratch. Vadret: VAST, 1 PB, home. Archive and backup: 2 x 130 PB tape libraries",
  "cooling": "Direct liquid cooling in HPE Cray EX cabinets (no cabinet fans, coolant distribution unit); facility cooled with water from Lake Lugano",
  "power_mw": "7.12 MW (7,124 kW, power measurement level 3, TOP500 June 2026 list)",
  "rmax_pflops": "434.90 PFlop/s (TOP500 June 2026 list)",
  "rpeak_pflops": "574.84 PFlop/s (TOP500 June 2026 list)",
  "status_note": "GH200 nodes installed January to June 2024 and officially inaugurated 14 September 2024; earlier partitions date from October 2020 (1,024 AMD Rome nodes); successor to Piz Daint, operated by CSCS for Swiss and international researchers"
 },
 "not_confirmed": "No source read names the maker of the HBM3, LPDDR5X or DDR4 memory, the NVMe and hard drives inside the ClusterStor and VAST systems, the tape libraries behind the 2 x 130 PB archive (CSCS names only HPE Data Management Framework as the software that moves data between tiers), the cabinet cold plates, quick disconnects, coolant distribution units and rectifiers, or the Slingshot cables and optics. Facility cooling has no named supplier either. CSCS says Lake Lugano water cools the machine, and a CSCS-authored article on the building describes lake water drawn from 45 m depth at about 6 C, up to 760 litres per second pumped (460 for CSCS), heat exchangers between the lake circuit and the internal circuit, a first circuit sized for supercomputers of up to 14 MW and a second mid-temperature circuit for a further 7 MW. That article describes the building's design rather than Alps's own load and does not name the pump, heat-exchanger or electrical vendors. The TOP500 figures cover the GH200 partition only; CSCS describes the full infrastructure as heterogeneous and still evolving, and its own documents give different storage and node-memory figures in places (see the comments on those lines).",
 "lines": [
  {
   "n": 1,
   "line": "System design and cabinets (HPE Cray EX, liquid-cooled; 24 GH200 cabinets, 8 chassis each, no cabinet fans, cooled by direct liquid cooling and a coolant distribution unit)",
   "parts": [
    {
     "id": "4.01",
     "name": "Rack-scale systems",
     "url": "/parts/servers-racks-and-integration/rack-scale-systems-nvl72-class-mi400-helios-class"
    }
   ],
   "companies": [
    {
     "id": "hpe",
     "name": "HPE",
     "url": "/companies/hpe",
     "reported": false,
     "note": "system integrator and cabinet designer (HPE Cray EX); four power shelves per cabinet with up to six 12.5 or 15 kW rectifiers each"
    }
   ],
   "unresolved": [
    "makers of the cold plates, quick disconnects, coolant distribution units and rectifiers (not named in any source read)"
   ],
   "sources": [
    "https://docs.cscs.ch/alps/hardware/",
    "https://www.cscs.ch/fileadmin/user_upload/contents_publications/press_releases/20210412_alps/20210412_CSCS_HPE_NVIDIA_Alps_EN.pdf"
   ]
  },
  {
   "n": 2,
   "line": "GH200 compute blade and node (2 nodes per blade, 4 Grace Hopper modules and 4 NICs per node; 7 of 8 blade slots filled per chassis; 2,688 nodes)",
   "parts": [
    {
     "id": "4.03",
     "name": "GPU baseboards & compute trays",
     "url": "/parts/servers-racks-and-integration/gpu-baseboards-and-compute-trays-hgx-bianca"
    }
   ],
   "companies": [
    {
     "id": "hpe",
     "name": "HPE",
     "url": "/companies/hpe",
     "reported": false,
     "note": "HPE Cray EX254n liquid-cooled blade (TOP500 system name)"
    },
    {
     "id": "nvidia",
     "name": "Nvidia",
     "url": "/companies/nvidia",
     "reported": false,
     "note": "Grace Hopper modules fitted four to a node"
    }
   ],
   "unresolved": [],
   "sources": [
    "https://docs.cscs.ch/alps/hardware/",
    "https://top500.org/system/180259/",
    "https://www.servethehome.com/8x-nvidia-grace-hopper-superchips-arm-in-a-blade-hpe-cray-ex254n-at-gtc-2024/"
   ]
  },
  {
   "n": 3,
   "line": "GH200 Grace Hopper Superchip (72-core Grace CPU plus Hopper GPU with 96 GB HBM3, joined by NVLink-C2C; 10,752 superchips)",
   "parts": [
    {
     "id": "1.01",
     "name": "Merchant GPUs / accelerators",
     "url": "/parts/compute-silicon-and-packaging/merchant-gpus-accelerators"
    },
    {
     "id": "1.05",
     "name": "Server CPUs",
     "url": "/parts/compute-silicon-and-packaging/server-cpus"
    }
   ],
   "companies": [
    {
     "id": "nvidia",
     "name": "Nvidia",
     "url": "/companies/nvidia",
     "reported": false,
     "note": null
    },
    {
     "id": "arm",
     "name": "Arm",
     "url": "/companies/arm",
     "reported": false,
     "note": "Neoverse V2 core design used in the Grace CPU"
    }
   ],
   "unresolved": [
    "which companies made the 96 GB HBM3 stacks and the 128 GB of LPDDR5X per module (CSCS, HPE and Nvidia do not name them)",
    "advanced-packaging flow and substrate suppliers for the GH200 module (none named in any source read)"
   ],
   "sources": [
    "https://www.cscs.ch/computers/alps",
    "https://developer.nvidia.com/blog/nvidia-grace-hopper-superchip-architecture-in-depth/"
   ]
  },
  {
   "n": 4,
   "line": "Hopper GPU die fabrication (TSMC 4N process)",
   "parts": [
    {
     "id": "1.09",
     "name": "Leading-edge foundry",
     "url": "/parts/compute-silicon-and-packaging/leading-edge-foundry"
    }
   ],
   "companies": [
    {
     "id": "tsmc",
     "name": "TSMC",
     "url": "/companies/tsmc",
     "reported": false,
     "note": "fabricates the GH100 die on its 4N process per Nvidia's Hopper write-up; CSCS's own pages do not name the foundry"
    }
   ],
   "unresolved": [
    "foundry for the Grace CPU die (not named in any source read)"
   ],
   "sources": [
    "https://developer.nvidia.com/blog/nvidia-hopper-architecture-in-depth/",
    "https://www.cscs.ch/computers/alps"
   ]
  },
  {
   "n": 5,
   "line": "Network interface cards (HPE Cassini, Slingshot-11 at 200 Gb/s; 4 per GH200 node, 2 per Rome node)",
   "parts": [
    {
     "id": "1.06",
     "name": "NICs / SuperNICs",
     "url": "/parts/compute-silicon-and-packaging/nics-supernics"
    }
   ],
   "companies": [
    {
     "id": "hpe",
     "name": "HPE",
     "url": "/companies/hpe",
     "reported": false,
     "note": "Cassini NIC in the Slingshot-11 interconnect"
    }
   ],
   "unresolved": [],
   "sources": [
    "https://docs.cscs.ch/alps/hardware/",
    "https://cpe.ext.hpe.com/docs/latest/getting_started/HPE-Cassini-Performance-Counters.html"
   ]
  },
  {
   "n": 6,
   "line": "Slingshot-11 fabric (64-port Rosetta switch ASIC on TSMC 16 nm; 200 Gb/s links; also carries Capstor, Iopsstor and Ritom storage traffic)",
   "parts": [
    {
     "id": "3.02",
     "name": "Scale-out switch ASICs",
     "url": "/parts/networking-and-interconnect/scale-out-switch-asics"
    },
    {
     "id": "3.03",
     "name": "Switch systems & white-box ODMs",
     "url": "/parts/networking-and-interconnect/switch-systems-and-white-box-odms"
    },
    {
     "id": "1.09",
     "name": "Leading-edge foundry",
     "url": "/parts/compute-silicon-and-packaging/leading-edge-foundry"
    }
   ],
   "companies": [
    {
     "id": "hpe",
     "name": "HPE",
     "url": "/companies/hpe",
     "reported": false,
     "note": "designs and supplies the Slingshot switches; Alps is an HPE Cray EX system built on Slingshot"
    },
    {
     "id": "tsmc",
     "name": "TSMC",
     "url": "/companies/tsmc",
     "reported": false,
     "note": "Rosetta switch ASIC process, per a paper co-authored by HPE engineers (the paper describes Slingshot generally, not Alps's switch count)"
    }
   ],
   "unresolved": [
    "switch counts, topology, and the cable and optics vendors for Alps's Slingshot links (not stated in any source read)"
   ],
   "sources": [
    "https://www.cscs.ch/computers/alps",
    "https://arxiv.org/html/2507.02404",
    "https://arxiv.org/abs/2008.08886",
    "https://docs.cscs.ch/alps/storage/"
   ]
  },
  {
   "n": 7,
   "line": "CPU-only partition (1,024 nodes of two AMD EPYC 7742 Rome CPUs, 64 cores each; installed from October 2020)",
   "parts": [
    {
     "id": "1.05",
     "name": "Server CPUs",
     "url": "/parts/compute-silicon-and-packaging/server-cpus"
    }
   ],
   "companies": [
    {
     "id": "amd",
     "name": "AMD",
     "url": "/companies/amd",
     "reported": false,
     "note": null
    }
   ],
   "unresolved": [
    "maker of the DDR4 DIMMs (256 GB or 512 GB per node; not named)"
   ],
   "sources": [
    "https://www.cscs.ch/computers/alps",
    "https://docs.cscs.ch/alps/hardware/"
   ]
  },
  {
   "n": 8,
   "line": "AMD Instinct MI300A partition (128 nodes, 4 APUs per node, 512 in total; HPE Cray EX255A blades)",
   "parts": [
    {
     "id": "1.01",
     "name": "Merchant GPUs / accelerators",
     "url": "/parts/compute-silicon-and-packaging/merchant-gpus-accelerators"
    },
    {
     "id": "1.05",
     "name": "Server CPUs",
     "url": "/parts/compute-silicon-and-packaging/server-cpus"
    },
    {
     "id": "1.09",
     "name": "Leading-edge foundry",
     "url": "/parts/compute-silicon-and-packaging/leading-edge-foundry"
    },
    {
     "id": "4.03",
     "name": "GPU baseboards & compute trays",
     "url": "/parts/servers-racks-and-integration/gpu-baseboards-and-compute-trays-hgx-bianca"
    }
   ],
   "companies": [
    {
     "id": "amd",
     "name": "AMD",
     "url": "/companies/amd",
     "reported": false,
     "note": "MI300A APU, 24 Zen 4 cores plus CDNA 3 GPU on one package"
    },
    {
     "id": "tsmc",
     "name": "TSMC",
     "url": "/companies/tsmc",
     "reported": false,
     "note": "AMD lists the MI300A lithography as TSMC 5nm | 6nm FinFET"
    },
    {
     "id": "hpe",
     "name": "HPE",
     "url": "/companies/hpe",
     "reported": false,
     "note": "EX255A (Parry Peak) blade, two nodes per blade"
    }
   ],
   "unresolved": [
    "maker of the HBM stacks on the MI300A (not named)"
   ],
   "sources": [
    "https://docs.cscs.ch/alps/hardware/",
    "https://www.cscs.ch/computers/alps",
    "https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html"
   ]
  },
  {
   "n": 9,
   "line": "NVIDIA A100 partition (144 nodes, one AMD EPYC host and 4 A100 GPUs per node, 576 GPUs) and AMD Instinct MI250X partition (24 nodes, 4 MI250X per node, 96 GPUs; HPE Cray EX235A blades)",
   "parts": [
    {
     "id": "1.01",
     "name": "Merchant GPUs / accelerators",
     "url": "/parts/compute-silicon-and-packaging/merchant-gpus-accelerators"
    },
    {
     "id": "4.03",
     "name": "GPU baseboards & compute trays",
     "url": "/parts/servers-racks-and-integration/gpu-baseboards-and-compute-trays-hgx-bianca"
    }
   ],
   "companies": [
    {
     "id": "nvidia",
     "name": "Nvidia",
     "url": "/companies/nvidia",
     "reported": false,
     "note": "A100 GPUs, 4 per node"
    },
    {
     "id": "amd",
     "name": "AMD",
     "url": "/companies/amd",
     "reported": false,
     "note": "MI250X GPUs and EPYC host CPUs (EPYC 7A53 Trento in the MI250X nodes)"
    },
    {
     "id": "hpe",
     "name": "HPE",
     "url": "/companies/hpe",
     "reported": false,
     "note": "EX235A (Bard Peak) blade for the MI250X nodes, two nodes per blade"
    }
   ],
   "unresolved": [
    "maker of the A100 blades (CSCS names only the \"Grizzly Peak\" codename; HPE is the system integrator but the blade is not itemised)"
   ],
   "sources": [
    "https://www.cscs.ch/computers/alps",
    "https://docs.cscs.ch/alps/hardware/"
   ]
  },
  {
   "n": 10,
   "line": "Capstor and Iopsstor Lustre file systems (HPE ClusterStor E1000D, 129 PB raw on 8,480 x 16 TB hard drives; ClusterStor E1000F, 7.2 PB raw on 240 x 30 TB NVMe drives)",
   "parts": [
    {
     "id": "2.10",
     "name": "AI storage systems",
     "url": "/parts/memory-and-storage/ai-storage-systems-parallel-file-systems-object"
    }
   ],
   "companies": [
    {
     "id": "hpe",
     "name": "HPE",
     "url": "/companies/hpe",
     "reported": false,
     "note": "ClusterStor E1000 hardware (Capstor is 80 object storage servers and 6 metadata servers, Iopsstor is 20 and 2); HPE Data Management Framework moves data between storage tiers"
    }
   ],
   "unresolved": [
    "hard-drive and NVMe SSD vendors inside the ClusterStor systems (not named)",
    "vendor of the 2 x 130 PB tape libraries (not named; CSCS names only HPE Data Management Framework as the tiering software)"
   ],
   "sources": [
    "https://docs.cscs.ch/alps/storage/",
    "https://www.cscs.ch/computers/alps"
   ]
  },
  {
   "n": 11,
   "line": "Ritom (VAST, 10 PB, 7.2 PB real, 528 x 15 TB SSD) and Vadret (VAST, 1 PB home storage) file systems",
   "parts": [
    {
     "id": "2.10",
     "name": "AI storage systems",
     "url": "/parts/memory-and-storage/ai-storage-systems-parallel-file-systems-object"
    }
   ],
   "companies": [
    {
     "id": "vast-data",
     "name": "VAST Data",
     "url": "/companies/vast-data",
     "reported": false,
     "note": "VAST Data Platform behind Ritom scratch and Vadret home; Ritom connects directly to the Slingshot fabric with no gateway layer, Vadret to CSCS's Ethernet network"
    }
   ],
   "unresolved": [
    "SSD vendor for the 528 x 15 TB drives (not named)"
   ],
   "sources": [
    "https://docs.cscs.ch/alps/storage/",
    "https://www.vastdata.com/press-releases/cscs-research-highlights-vast-datas-role-in-ai-and-hpc-infrastructure"
   ]
  },
  {
   "n": 12,
   "line": "System management (HPE Cray System Management microservices for node configuration, imaging and grouping; Slurm and vCluster layers on top)",
   "parts": [
    {
     "id": "11.02",
     "name": "Cluster management & GPU orchestration",
     "url": "/parts/software-tools-and-services-touching-the-hardware-optional/cluster-management-and-gpu-orchestration-out-of-scope-for-a-part"
    }
   ],
   "companies": [
    {
     "id": "hpe",
     "name": "HPE",
     "url": "/companies/hpe",
     "reported": false,
     "note": "Cray System Management (CSM) is the management plane for Alps"
    }
   ],
   "unresolved": [],
   "sources": [
    "https://arxiv.org/html/2507.02404",
    "https://top500.org/system/180259/"
   ]
  }
 ]
}