The Nobel Prize-Winning Supercomputer Designed to Accelerate Science
Dell Technologies will build NERSC 10, the next flagship supercomputer of the National Energy Research Scientific Computing Centre (NERSC), under a new DOE contract. DOE user facility NERSC is at Berkeley Lab, also known as Lawrence Berkeley National Laboratory. While visiting Berkeley Lab, U.S. Secretary of Energy Chris Wright made the news.
The 10th-generation computer system NERSC 10 is housed in the National Energy Research Scientific Computer Centre (NERSC), a DOE Office of Science user facility. HPC facilities at NERSC enable scientific research in physics, climate science, energy research, biology, and materials science.
Overview of NERSC 10
- NERSC 10 will replace Perlmutter, NERSC’s premier supercomputer, in 2024.
- The deployment of NERSC 10 is expected to occur in 2026 or later.
- Developed to meet DOE Office of Science’s rising data and computing needs until 2030.
- The Lawrence Berkeley National Laboratory (LBNL) in California will host NERSC 10 again.
Key features
NERSC 10 Extreme Performance’s main objectives are:
- Computer power at the exascale or near-exascale level.
- Support for a variety of workloads that are getting more complicated.
Architecture in Balance:
- High capacity and bandwidth of memory.
- I/O systems for fast storage.
- Sophisticated interconnects for communication with minimal latency.
Integration of AI with HPC:
- In addition to standard simulations, AI and machine learning (ML) workloads are supported.
- designed with hybrid workflows in mind.
Efficiency of Energy:
- Aiming for increased FLOP (floating point operation) power efficiency.
- Possible investigation of green technology such as liquid cooling.
User-focused design:
- Continuity with the current user experience and software stack of NERSC.
- A focus on usability and productivity for a wide range of scientists.
Technical Expectations (Speculative but Informed)
While official technical specifications haven’t been finalized publicly, NERSC 10 is expected to include:
| Feature | Expected Details |
|---|
| Performance | >1 ExaFLOP (potentially rivaling DOE’s Aurora or Frontier systems) |
| Compute Architecture | Likely to feature heterogeneous compute nodes: CPUs + GPUs or accelerators |
| Vendors | Could be from familiar HPC players like HPE/Cray, AMD, NVIDIA, or Intel |
| Memory | Multi-level memory hierarchy, potentially including HBM and persistent memory |
| Storage | Fast parallel storage, likely using Lustre or newer file systems with SSD tiers |
| Networking | High-speed interconnects (e.g., Cray Slingshot, InfiniBand) |
| Software Support | Continuation of NERSC’s Shifter containers, science gateways, and data tools |
Purchasing and Development
- DOE usually releases a Request for Proposal (RFP) a number of years prior to the delivery of the system.
- Community Involvement: To make sure the system satisfies practical research needs, NERSC solicits feedback from the scientific user community throughout the system design phase.
Strategic Significance
- Science Impact: Assists hundreds of research projects and more than 9,000 users in all DOE mission areas.
- Leadership Role: Unlike some other more experimental exascale systems, NERSC systems are made to be extremely useable and accessible to a wide range of scientists.
Due in 2026, the new system will be called “Doudna” in honour of biologist Jennifer Doudna of Berkeley Lab, who won the 2020 Nobel Prize for Chemistry for her pioneering discovery on the gene-editing technique CRISPR. Secretary Wright, who acknowledged Doudna’s biological findings and the promise for computational capabilities to accelerate progress in areas like curing diseases and tumours, expressed his astonishment and happiness at the nomination.
NVIDIA’s next-generation Vera Rubin platform will power the Dell Technologies system that will house the Doudna supercomputer. Large-scale high-performance computing (HPC) workloads, including high-energy physics, molecular dynamics, and AI training and inference, are among the tasks it is being designed to serve. By combining simulation, data, and artificial intelligence into a single, seamless platform, it is intended to offer a stable environment for the workflows that support cutting-edge science and hasten discoveries.
The system “represents DOE’s commitment to advancing American leadership in science, AI, and high-performance computing,” according to Secretary Wright. According to him, Doudna is a “powerhouse for rapid innovation” that will revolutionise attempts to advance quantum computing breakthroughs and provide plentiful, reasonably priced energy. Wright compared AI to “the Manhattan Project of the time,” emphasizing that Doudna will help ensure that American scientists receive the resources they need to prevail in the global competition for supremacy in AI.
NERSC Director Sudip Dosanjh said the Doudna supercomputer will speed up several scientific processes. In order to get its 11,000 users ready to take use of the system’s enhanced workflow capabilities, NERSC is working with NVIDIA and Dell. Through the Energy Sciences Network (ESnet), Doudna will be linked to DOE observational and experimental facilities, enabling scientists to analyse data in almost real-time and stream it with ease. As a result of this integration, the supercomputer is no longer merely a passive player in the workflow.
Doudna is anticipated to be crucial in promoting innovations in a number of industries. Since NERSC funds fusion research, Doudna is well-positioned to accelerate the discovery of abundant, usable energy. Its powerful GPUs will allow DOE-funded researchers to quickly integrate large-scale Artificial Intelligence into their workflows, speeding up basic physics, biomolecular modelling, and advanced materials design research. Modern quantum simulation tools, such as NVIDIA’s CUDA-Q platform, will also be supported by the system enabling the co-design of next integrated quantum-HPC systems and the creation of scalable quantum algorithms.
According to NVIDIA, the Vera-Rubin CPU-GPU platform and Dell’s most cutting-edge ORv3 direct liquid-cooled server technologies are used. It will have a fast NVIDIA Quantum-X800 InfiniBand networking architecture, be powered by Dell Integrated Rack Scalable Systems and PowerEdge servers with NVIDIA accelerators, and come with high-performance data management and storage options.
In terms of scientific output, Doudna is anticipated to outperform NERSC’s current flagship supercomputer, Perlmutter, by more than ten times. It is predicted to utilise just two to three times the power of Perlmutter, which will boost performance by three to five times per watt. The ultimate objective is to drastically cut down on the amount of time required for important scientific breakthroughs.
“We’re not just building a faster computer,” said Nick Wright, NERSC’s Doudna main architect and group lead for advanced technologies. We’re developing a framework that enables researchers to think more broadly and make discoveries more quickly. He went on to say that the system is made to address global issues quickly and that it is intended to spur research in physics, chemistry, and other unimagined domains.
