Supercomputers/Systems/Alps
Switzerland · operational · mixed
Alps
Deep dive: Isambard-AI versus Alps: two GH200 routes
Operated by CSCS at CSCS, Lugano .
Researched with AI assistance, checked by scripts. These rows were researched by AI agents reading public operator, laboratory and vendor sources, appended by scripts and checked by automated validation; no subject-matter expert has reviewed them (how this site is made). A mechanical second-reader pass has since fetched every cited source: 1350 of 1581 systems have a readable citation that names them, and 64 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
- 434.9 PFlop/s
- Rpeak
- 574.8 PFlop/s
- Rmax ÷ Rpeak
- 75.7%
- Cores
- 2,121,600
- Power
- 7.12 MW
- Per watt
- 61.0 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.
How it compares
Ranked against the other machines in this dataset that report the same figure: those that first came into service within two years of Alps, and those in Switzerland. Computed from our tables; a metric is ranked only when at least eight peers report it.
| Metric | Value | Against its era | Against its country |
|---|---|---|---|
| Rmax | 434.9 PFlop/s | #9 of 234 · top 3% | #1 of 11 · highest |
| Energy efficiency | 61.0 GF/W | #23 of 121 · top 18% | - |
| Rmax ÷ Rpeak | 75.7 % | #108 of 231 · top 47% | #6 of 11 · top 50% |
| Power draw | 7,124 kW | #12 of 126 · top 9% | - |
Every metric is ranked largest-first, so a high power-draw rank means a big power draw, not a good one. Rmax ÷ Rpeak uses HPL-measured results only. Rankings move as machines are added; they describe this dataset, not the world.
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 | - HPE Cray EX254n | Hewlett Packard Enterprise | - | Reported | cscs.chnvidia.com |
| Accelerator | NVIDIA GH200 Grace Hopper Superchip Accelerators Arm CPU + CUDA GPU · TSMC 4N NVIDIA GH200 Superchip, 4 per node | NVIDIA | 10,752 reported | Reported | cscs.chnvidia.com |
| CPU | NVIDIA Grace CPUs Armv9 (Neoverse V2) · 72C · 3.1 GHz · TSMC 4N NVIDIA Grace 72C (in GH200) | NVIDIA | - | Reported | cscs.chnvidia.com |
| Interconnect | HPE Slingshot-11 Interconnect Dragonfly · 200 Gb/s per port · Around 2 microseconds Slingshot-11 | Hewlett Packard Enterprise | - | Reported | cscs.chnvidia.com |
| Storage | HPE Cray ClusterStor E1000 Storage Lustre appliance, flash and disk tiers · 700 PB on Frontier (Orion) · About 10 TB/s peak Lustre on ClusterStor | Hewlett Packard Enterprise | - | Reported | cscs.chnvidia.com |
| Cooling | Direct liquid cooling, warm water Cooling Direct-to-chip cold plates, facility water loop · Removes 90+ percent of node heat to liquid direct liquid cooling | Hewlett Packard Enterprise | - | Reported | cscs.chnvidia.com |
| Operating system | HPE Cray Operating System Operating systems SUSE-derived Linux, container-based management · Proprietary layer over open-source Linux HPE Cray Operating System | Hewlett Packard Enterprise | - | Reported | cscs.chnvidia.com |
Engineering, funding and access
What Alps runs on, who paid for it and how people get time on it. Each line carries who said it and when.
Hardware
| Item | Detail | As of | Basis | Source |
|---|---|---|---|---|
| Compute nodes | 2,688 nodesNVIDIA Grace Hopper nodes; Alps is heterogeneous and also holds 1,024 AMD EPYC Rome CPU nodes, 144 A100 nodes, 128 MI300A nodes and 24 MI250X nodes | 2024 | Operator stated | cscs.ch |
| Storage capacity | 110 PBScratch on hard disk, 100 PB plus 10 PB; also 5 PB plus 1 PB SSD scratch and two 130 PB tape libraries for archive and backup | 2024 | Operator stated | cscs.ch |
| Interconnect and topology | HPE Cray Slingshot-11, 200 Gb/s injection bandwidth per module | 2024 | Operator stated | cscs.ch |
Software
| Item | Detail | As of | Basis | Source |
|---|---|---|---|---|
| Programming and software stack | vClusters: software-defined clusters carved from Alps, with tenants such as MeteoSwiss, the CSCS User Lab and machine learning and AI platforms | 2024 | Operator stated | cscs.ch |
Access
| Item | Detail | As of | Basis | Source |
|---|---|---|---|---|
| How researchers get time | CSCS User Lab and tenant vClusters; open to researchers in Switzerland and worldwide | 2024 | Operator stated | cscs.ch |
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 | 10,752 accelerators | 2026-10 | Operator stated | GH200 superchips (2,688 nodes x 4), per CSCS and the Swiss AI Apertus report. Excludes the A100, MI250X and MI300A partitions. | cscs.charxiv.org |
| PUE | 1.2 | 2022-01 | Operator stated | CSCS states PUE 'below 1.2' for its Lugano data centre (lake-water cooling, 14 MW first circuit). Facility-wide, from a 2022 article, so an upper bound. | cscs.ch |
The read
The first large European system built on NVIDIA's Grace Hopper superchip, which means NVIDIA occupies the CPU role and the accelerator role simultaneously: the same structural pattern as El Capitan, arrived at from the opposite direction. Alps is also explicitly built as a multi-tenant research infrastructure rather than a single-program machine, which is why its workload class is mixed rather than hpc.
Timeline
- 2020-10 First Alps cabinets installed with 1,024 AMD EPYC nodes and 10 PB scratch source
- 2021 Announced source
- 2023-07 Alps scratch storage expanded to 100 PB with 1 TB/s throughput and HPE Data Management Framework enabled source
- 2024-01 Stepwise installation of 2,688 NVIDIA Grace Hopper nodes began, completed by June 2024 source
- 2024-03 First operational source
- 2024-06 List debut at #6 (June 2024) source
- 2024-09 Alps officially inaugurated source
Public list appearances
Pointers only: rank and edition, linking to the canonical entry. We do not reproduce list tables. See sourcing policy.
Site & facility
CSCS, Lugano
- Location
- Lugano, Switzerland
- Cooling
- Lake-water cooling (Lake Lugano)
- Commissioned
- 2012
Change history
- 2026-10-04 Added AI datacenter metrics for alps
- 2026-09-22 Gained a Green500 rank pointer (#14 on the 2024-11 list)
Nearby in scale
The machines closest to Alps by Rmax, smaller to larger.
Related systems
Same site
- Piz Daint · Switzerland, decommissioned, 2017
- Blue Brain IV · Switzerland, decommissioned, 2013
Same operator
- Piz Kesch · Switzerland, decommissioned, 2015
- Todi · Switzerland, decommissioned, 2011
- Monte Rosa · Switzerland, decommissioned, 2009
- CSCS NEC SX-3/22 · Switzerland, operational
Same accelerator
- CEA-HE · France, operational, 2024
- Helios · Poland, operational, 2024
- Isambard-AI · United Kingdom, operational, 2024
- JEDI · Germany, operational, 2024
- JETI · Germany, operational, 2024
- ROMEO-2025 · France, operational, 2024
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 |
|---|---|---|
| cscs.ch | names system, part and count | NVIDIA Grace, HPE Slingshot-11, HPE Cray Operating System, HPE Cray ClusterStor E1000 · counts: 10752 |
| nvidia.com | blocked our fetch | - |
Checked 2026-10-18 by pnpm verify. Re-run it and the table changes with the web.
Related news
- ETH Zurich and EPFL release Apertus, a fully open LLM trained on Alps · ETH ZurichApertus used more than 10 million GPU hours on Alps to train 8B and 70B parameter models with open weights, data and training code, a rare case in this dataset of a system's compute being fully accounted for by one named downstream model rather than a mix of undisclosed workloads.
- Piz Daint powered off after 13 years, cleared for Alps · CSCSCSCS retired the Cray XC50/XC40 system that had anchored Swiss GPU computing since 2012, closing out the last pre Grace Hopper system in this dataset's Swiss lineage and completing the multi year handover to Alps that began with Alps' 2024 inauguration.
Further reading
Inside a node
Why accelerators exist, what the memory hierarchy costs, and how the CPU and accelerator merged onto one package.
Power and cooling
Why air ran out at 20 kW a rack, how warm-water cooling removes the chillers, and what PUE conceals.
Operating a supercomputer
Acceptance testing, why "installed" and "in production" are six months apart, and the machine lifecycle.