The Lawrence Berkeley National Laboratory and NVIDIA introduce a groundbreaking platform for science powered by artificial intelligence.
The race for global scientific supremacy takes a new step with the announcement of the supercomputer Doudna, also known as NERSC-10, being built at the Lawrence Berkeley National Laboratory (USA). This high-performance computing system, driven by NVIDIA’s Vera Rubin architecture and based on Dell infrastructure, is designed to lead a new era of AI-accelerated scientific discoveries.
Named in honor of Jennifer Doudna, Nobel laureate and pioneer of CRISPR technology, the new supercomputer represents a strategic effort by the U.S. Department of Energy (DOE) to position the country at the forefront of scientific research in key areas such as energy, medicine, astronomy, and particle physics.
An Investment for the Future of Science
“The construction of Doudna reaffirms the DOE’s commitment to U.S. scientific leadership in artificial intelligence and advanced computing,” said Chris Wright, Secretary of Energy. NVIDIA emphasized the disruptive nature of the project: “Doudna is a time machine for science, capable of compressing years of research into just days,” stated Jensen Huang, founder and CEO of NVIDIA.
Beyond sheer performance, the supercomputer has been designed to integrate directly into scientific workflows. This means it will be able to receive and process real-time data from telescopes, particle detectors, or genomic sequencers, all thanks to the ESnet network and low-latency connectivity with NVIDIA Quantum-X800 InfiniBand.
An Exponential Leap in Efficiency and Power
The new system is set to surpass its predecessor, Perlmutter, by more than 10 times in scientific discovery capacity, using only two to three times more energy. This represents an improvement of three to five times in performance per watt, thanks to innovations in chip design, dynamic load balancing, and system-level efficiency.
High-Impact Scientific Applications
Among the fields where Doudna is expected to make a significant difference are:
- Nuclear Fusion: Advanced simulations to accelerate the acquisition of clean energy.
- Material Science: AI models that help design new superconducting materials.
- Drug Discovery: Ultra-fast workflows for protein folding in preparation for future pandemics.
- Astronomy: Real-time processing of data from the DESI spectroscopic instrument to map the universe.
Over 11,000 researchers will benefit from its near-instantaneous capabilities, allowing for more complex questions and answers at unprecedented speed.
Artificial Intelligence at the Service of Science
Doudna has been optimized to run workloads that combine traditional HPC, AI, real-time data streams, and even quantum algorithms. Platforms like NVIDIA CUDA-Q will enable the scalable development of quantum algorithms and integration with classical HPC systems.
Currently, more than 20 scientific teams are adapting their workflows to the new system through the NERSC Science Acceleration Program. These include initiatives in particle physics, climate, molecular biology, and computational chemistry.
From Protein Prediction to Fundamental Physics
The supercomputer will also be key in advancing AI models already recognized with a Nobel Prize, such as those developed by David Baker to predict protein structures. Other researchers, like Benjamin Nachman, use AI to enhance data analysis in particle physics, and collaborations like Open Molecules 2025, in partnership with Meta, aim to model complex chemical reactions from vast data.
Infrastructure Ready for What’s Next
On a technical level, Doudna relies on the NVIDIA Vera Rubin platform, which combines high-performance CPUs with coherent GPUs, allowing for more efficient data sharing among processors. Compatible tools include PyTorch, TensorFlow, cuDNN, Holoscan, and CUDA-Q, optimized for Rubin’s NVLink architecture.
This new generation of supercomputing is not a luxury but a necessity. As Nick Wright, lead architect of Doudna at NERSC, concludes: “We no longer think of supercomputers as isolated tools in a corner of the lab. They are now an integral part of the discovery cycle, directly connected to experiments, telescopes, and sensors.”
The launch of Doudna is expected for 2026. From that point on, it is anticipated to become a cornerstone for addressing some of the most complex questions of the 21st century.
Source: NVIDIA Blogs