The road to deploying the next flagship supercomputer at the National Energy Research Scientific Computing Center (NERSC) reached an important milestone in January with the delivery of the NERSC-10 Pilot Early Access System (EAS). Smaller than the full-size system, the EAS gives NERSC staff and the system vendor, Dell Technologies, the opportunity to refine assembly, delivery, installation, and integration processes, paving the way for seamless deployment of the full-scale production system in late 2026. The EAS is currently being installed at NERSC, a U.S. Department of Energy (DOE) user facility located at Lawrence Berkeley National Laboratory (Berkeley Lab).
"We are excited to take delivery of the early access system," said NERSC Director Sudip Dosanjh. "This is an important step in deploying Doudna for NERSC's 11,000-plus users. We plan to use the EAS to develop and test the software stack for Doudna in collaboration with Dell Technologies, NVIDIA, and our other partners. This opportunity to polish our processes helps ensure a great user experience when we deploy the full system."
"NERSC is where the future of science becomes real," said Paul Perez, Senior Fellow, Dell Technologies Federal. "The early access system is the first step toward Doudna, which will set the blueprint for how federal agencies scale HPC and AI - securely, efficiently, and at mission speed."
A stepping stone to Doudna
The next NERSC supercomputer will be called Doudna after Jennifer Doudna, the biochemist honored with a Nobel Prize in recognition of her work on the gene-editing technology CRISPR. The EAS will be called Cech (pronounced "check") in honor of Thomas Cech, the chemist awarded a Nobel Prize in Chemistry for his discovery of the catalytic properties of RNA. Cech's work was a crucial stepping stone to Doudna's research, making him a natural choice for the EAS's namesake.
Cech spent time at Berkeley Lab as a summer intern in 1969 and then earned his Ph.D. in chemistry from the University of California, Berkeley in 1975. He later joined the faculty of the University of Colorado Boulder, where he became a Howard Hughes Medical Institute (HHMI) investigator in 1988 and Distinguished Professor of Chemistry and Biochemistry in 1990. In 1982 Cech and his research group announced that an RNA molecule from the pond animal Tetrahymena could catalyze biochemical reactions, the first exception to the long-held belief that only proteins could act as enzymes. This work led to the 1989 Nobel Prize in Chemistry, shared with Sidney Altman of Yale University.
"Like supercomputers, each scientific discovery builds on the previous generation, and requires a community effort to lay the foundation for the next. Naming the EAS Cech is a great reminder of the connections between us and the impact of our contributions to the next generation," said Rollin Thomas, NERSC-10 Project Director.
Testing systems and technologies
Deploying a new supercomputer is a complex process, and the EAS allows NERSC and Dell to test the entire process, from delivery and installation to software setup, before the full system arrives. This early hands‑on work ensures the final supercomputer integrates smoothly into NERSC's operations.
Moreover, many of the technologies built into Cech are new: designed by Dell, powered by NVIDIA AI infrastructure, and operated by NERSC, with extreme-scale storage and data services provided by VAST Data, Cech is an example of a public-private partnership to deploy cutting-edge technologies in DOE's state-of-the-art data centers. Cech includes 72 NVIDIA Grace CPUs and 144 NVIDIA Blackwell GPUs - the precursors to the Vera-Rubin platform planned for the Doudna system - interconnected by the low-latency, high-throughput NVIDIA Quantum-X800 InfiniBand networking platform - within Dell's energy-efficient IR7000-series direct liquid-cooled rack and the first U.S. deployment of Rittal's in-row V3.5 coolant distribution unit. Cech will also feature one of the first deployments of Dell's PowerCool Enclosed Rear Door Heat Exchanger (eRDHx), which is built to reduce cooling costs by 60% compared to traditional options. All these innovations hold enormous promise for users and for HPC in science.
"NERSC has a long history of deploying first-of-a-kind systems while pushing the boundaries of energy-efficient computing, and Cech is the latest example of this," said NERSC-10 Deputy Director for Systems Brian Friesen. "Integrating these cutting-edge NVIDIA AI infrastructure with Dell's pioneering liquid-cooling technology gives NERSC staff an incredible sandbox to test this powerful combination, and ensure a seamless, sustainable deployment for Doudna."
Cech uses the VAST AI OS to deliver both high-performance scratch and quality-of-service (QoS) data services, accommodating the distinct I/O demands of traditional modeling and simulation alongside AI-intensive workflows. Sustaining utilization of GPU-accelerated systems at this scale requires an architecture designed to deliver consistent throughput and service differentiation under highly mixed workloads.
In itself, Cech provides 5.76 petaflop/s FP64 computing power for simulation and 1.44 exaflop/s NVFP4, making it a powerful AI system that will be used to support the Genesis Mission, the DOE's AI initiative, after this initial development period.
"It's amazing to see the systems progress that's been made in recent years," said Dosanjh. "A single cabinet of NVIDIA Grace-Blackwell nodes has as much FP64 computing capability as twice the entire Edison system, NERSC's flagship system from 2014 through 2019, which filled the entire data-center floor."
In addition to the new hardware, NERSC staff will use the EAS period for benchmarking and to test a slew of new software systems for Doudna. This includes the complex modular software stack based on Omnia and OpenCHAMI needed to orchestrate and manage a supercomputer and its user software environment. NERSC and Dell will use Cech to collaborate on monitoring, telemetry, and alerting software infrastructure for Doudna and future systems. Cech will also be used to identify the best ways to make decades of data stored on NERSC's Community File System and HPSS archive available on Doudna, and to connect to external resources through ESnet, DOE's dedicated network for science.
All together, these test processes are key to a successful deployment, paving the way for Doudna just as Thomas Cech's research paved the way for Jennifer Doudna's. And when Doudna arrives, NERSC will be ready.
"Getting Cech in the data center now is essential to ensure the successful deployment of Doudna coming up at the end of the year," said Thomas. "Since the priority is for NERSC and Dell to learn from the EAS process, there isn't an immediate opportunity for users to get on the system. We hope users appreciate this period of intense preparation that will make deployment of Doudna as smooth as possible and make it a useful tool in the future as it supports the Genesis Mission and other groundbreaking research."
