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

Doudna (NERSC-10): every supplier in the machine

National Energy Research Scientific Computing Center, Lawrence Berkeley National Laboratory, United States

Accelerator
NVIDIA Rubin GPUs (Vera Rubin platform); quantity and memory not published
Host CPU
NVIDIA Vera (Vera Rubin platform); quantity not published
Integrator
Dell Technologies (Dell Integrated Rack Scalable Systems and PowerEdge servers)
Nodes
Not published. The early-access precursor Cech has 72 NVIDIA Grace CPUs and 144 NVIDIA Blackwell GPUs
Memory
Not published
Interconnect
NVIDIA Quantum-X800 InfiniBand, connected to DOE experiments and observatories over ESnet
Storage
Not published for the full system. NERSC's chief architect describes two tiers (a traditional parallel file system and a quality-of-service tier); the early-access system Cech uses VAST Data's storage
Cooling
Dell ORv3 direct-liquid-cooled servers. Cech: Dell IR7000-series direct-liquid-cooled rack, Rittal in-row V3.5 coolant distribution unit, Dell PowerCool enclosed rear-door heat exchanger
Power
Not published. NVIDIA says Doudna is expected to use 2-3 times Perlmutter's power for more than 10 times its scientific output
Rmax (HPL)
Not published; not yet on the TOP500
Rpeak
Not published. NERSC expects at least 10 times Perlmutter's performance
Status
Contract announced 29 May 2025; early-access system Cech delivered ahead of March 2026 and being installed at NERSC; the full system is scheduled for late 2026 (NERSC), with NVIDIA still describing it as being built in June 2026

Unmarked companies come from the operator's or the Department of Energy's own pages (energy.gov, nersc.gov, lbl.gov) or the vendor's own announcement (nvidia.com); reported companies come from named trade press whose article body was read in full, not just the headline. No company on this page is marked reported. This is a map of who plays where, not a confirmed bill of materials: DOE and NVIDIA announcements of 29 May 2025, NERSC's Doudna system page, Berkeley Lab's 17 March 2026 release on the Cech early-access system, NVIDIA's ISC 2026 release (22 June 2026), and a VAST Data blog interview with NERSC's chief architect (2 June 2025), researched 2026-10-22

This page is the parts-first view. For Doudna (NERSC-10)'s full system record on the sister site Compute Atlas, an open dataset of supercomputer and AI-cluster supply chains, see Doudna (NERSC-10) on Compute Atlas.

  1. System design, integration and rack platform (Dell Integrated Rack Scalable Systems with PowerEdge servers)
    Rack-scale systems 4.01
    Dell (prime contractor for NERSC-10 and system builder; the DOE announcement names the contract with Dell) · node count, rack count and rack power density for the full system — not published
    Sources: energy.gov · nersc.gov
  2. Compute platform (NVIDIA Vera Rubin CPU-GPU platform: Rubin GPUs and Vera CPUs)
    Merchant GPUs / accelerators 1.01 · Server CPUs 1.05
    Nvidia (Rubin GPUs and Vera CPUs; NVIDIA's June 2026 release names Doudna as a Dell system powered by Vera Rubin) · number of Rubin GPUs and Vera CPUs, and the HBM and LPDDR memory capacity — not published, maker of the HBM and LPDDR memory, and the foundry and packaging flow for the Rubin and Vera dies — no source read names them for Doudna
    Sources: nersc.gov · nvidianews.nvidia.com
  3. InfiniBand network (NVIDIA Quantum-X800, with quality-of-service for data streamed in over ESnet)
    Scale-out switch ASICs 3.02 · Switch systems & white-box ODMs 3.03
    Nvidia (Quantum-X800 InfiniBand networking platform) · switch counts, topology, and optical and copper cable vendors — not published
    Sources: nersc.gov · blogs.nvidia.com
  4. Direct liquid cooling (Dell ORv3 direct-liquid-cooled server technology)
    Direct-to-chip cold plates 5.01
    Dell (ORv3 direct-liquid-cooled server technology) · coolant distribution units, rear-door heat exchangers and facility plant for the full system — not published (see the Cech line for the early-access system), upgrades to the building's power and cooling (NERSC's chief architect says the building is being upgraded to carry more power) — no contractor or equipment maker named in any source read
    Sources: nersc.gov · vastdata.com
  5. Early-access system Cech (72 Grace CPUs, 144 Blackwell GPUs, Quantum-X800 InfiniBand; Dell IR7000-series direct-liquid-cooled rack with an in-row coolant distribution unit and rear-door heat exchanger)
    Rack-scale systems 4.01 · Coolant distribution units 5.05 · Rear-door heat exchangers 5.07
    Dell (designed Cech; IR7000-series rack and PowerCool enclosed rear-door heat exchanger), Nvidia (Grace CPUs, Blackwell GPUs and Quantum-X800 InfiniBand), Rittal (in-row V3.5 coolant distribution unit, the first U.S. deployment of that model) · whether the Doudna production racks will use the same Rittal CDU or the same rear-door design — not stated
    Sources: cs.lbl.gov
  6. Early-access system Cech storage (VAST AI OS: high-performance scratch and quality-of-service data services)
    AI storage systems 2.10
    VAST Data (storage and data services for Cech; not confirmed for the full Doudna system) · the storage system, vendor and capacity behind the full Doudna system — not published (a VAST-blog interview describes a traditional parallel file system plus a second tier but names no product for Doudna), maker of the flash and disk drives — not named in any source read
    Sources: cs.lbl.gov · vastdata.com

This is a planned system. DOE, NERSC, Dell and NVIDIA name Dell as the system builder and NVIDIA's Vera Rubin platform and Quantum-X800 InfiniBand as the compute and network, and say the full machine is due in late 2026, but none of them publishes a node count, GPU or CPU count, memory capacity, storage size or power draw for it. NVIDIA's 2025 blog says it is expected to use two to three times Perlmutter's power for more than ten times the scientific output; no megawatt figure appears in any source read, and NERSC's chief architect says the building is being upgraded for more power. Which storage system sits behind Doudna is not settled in the sources read. VAST Data supplies Cech, and the interview says a traditional parallel file system will sit beside a newer quality-of-service tier, without naming either product for Doudna. The coolant distribution units, rear-door heat exchangers and Cech's VAST storage are Cech facts. The lines will need revisiting once the full system is delivered. 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 · LUMI · Leonardo · Isambard-AI · Venado · Perlmutter · Eagle (Microsoft Azure) · Eos · Fugaku · MareNostrum 5 · Horizon. See also the AI rack teardowns and thesupercomputer overview.