On May 29, 2025, the U.S. Department of Energy (DOE) officially announced a partnership with Dell Technologies and NVIDIA to deploy Doudna, a next-generation supercomputing system, at Lawrence Berkeley National Laboratory (LBNL). The system is based on NVIDIA's next-generation Vera Rubin platform and is expected to be officially operational in 2026. This marks the beginning of a new era of intelligent convergence and heterogeneous architecture-driven scientific computing platforms in the United States, and also indicates that the technology infrastructure will set off a new round of competition on a global scale.
Ⅰ The signal behind the naming: a tribute to scientific innovation in the name of scientists
Named after Professor Jennifer Doudna, a renowned molecular biologist at the Berkeley Laboratory and a pioneer of CRISPR gene editing technology, Doudna not only continues the U.S. Department of Energy's tradition of naming supercomputers after outstanding scientists (such as Saul Perlmutter), but also demonstrates the importance of basic research to promote the scientific and technological revolution. As one of the founders of CRISPR technology, Professor Doudna's name symbolizes the scientific vision of "reshaping the world at the molecular level", which echoes the system's positioning for future research frontiers such as life sciences, quantum simulation, and climate modeling.
Ⅱ Technology Deconstructed: The convergence of the Vera Rubin platform with Dell's liquid cooling system
Doudna will be one of the world's first large-scale supercomputing systems built on NVIDIA's Vera Rubin platform. The platform integrates the heterogeneous computing architecture of NVIDIA Grace CPU+Blackwell GPU, combined with high-bandwidth memory (HBM3e) and NVLink interconnection technology, supports unified memory access, and provides an optimized path for AI training, scientific simulation, and quantum algorithm simulation.
The deployment is based on the Dell PowerEdge XE9640 series of high-density servers, equipped with Open RAN v3 (ORv3) direct liquid cooling architecture, supporting rack-level heat recovery and dynamic thermal management. Doudna's liquid cooling solution can reduce PUE (Power Usage Effectiveness) to as low as 1.08 compared to traditional air-cooled systems and reduce the carbon intensity of the entire system by approximately 30%.
Preliminary estimates show that Doudna's peak computing power may exceed 1 ExaFLOPS (exascale floating point operations), which is 10\~15 times that of the Perlmutter system currently running by LBNL (with a peak of 70.9 PFLOPS), which will help the United States officially enter the Exascale era.
Pictured: The U.S. Department of Energy and Dell/Nvidia are betting on Vera Rubin to build the Doudna supercomputer
Ⅲ Multidisciplinary research engine: Doudna's strategic application matrix
Doudna is not only a supercomputing platform, but also an experimental field for scientific research paradigms under the heterogeneous fusion architecture. Its three core research pillars include:
1. Materials Science and Molecular Modeling
By accelerating density functional theory (DFT) calculations and molecular dynamics simulations (MD), Doudna can compress the screening cycle for new materials from "months" to "days" or even "hours". For example, in solid-state battery development, Doudna can simulate more than 10,000 lithium-metal interface reaction models in parallel, significantly increasing the probability of material breakthroughs.
2. Quantum algorithm and simulation platform
Relying on the CUDA-Q development framework, Doudna supports native quantum circuit simulation, quantum noise modeling, variational quantum algorithm (VQA) training, etc., and has become a test field for building quantum-classical collaborative architectures. It is expected to complete 1 billion iterations of QPU state vectors per second, laying the foundation for the future deployment of quantum coprocessing units (QPUs).
3. AI research and basic model training
Doudna can deploy hundreds of billions of parameter-level AI basic models, playing a key role in biological sequence prediction, drug screening, protein structure modeling and other fields. It is estimated that the platform can complete the training task of AlphaFold scale models within 24 hours, and the training efficiency is about 4 times higher than that of the current mainstream platforms.
Ⅳ The leap of scientific research paradigm: from computing power to intelligent push
In the era of global scientific research gradually moving towards the era of "data flood", traditional HPC systems can no longer meet the scientific research needs of "high complexity + interdisciplinarity". The Vera Rubin platform, represented by Doudna, integrates AI computing power, scientific simulation, and large model inference capabilities, opening the transition from a "data-driven" to a "knowledge-driven" scientific research paradigm.
Taking climate simulation as an example, Doudna can make 100-year climate predictions at the resolution of a global 10-kilometer grid, and integrate AI models to identify and risk extreme climate events (such as hurricane paths) in advance. Compared with the current system, its calculation period can be shortened by more than 60%, which is of great significance for policy formulation and emergency response.
Ⅴ Strategic and geopolitical games: another card for the United States to consolidate its dominance in HPC
The launch of Doudna is the latest achievement of the DOE's "Exascale Computing Initiative (ECI)". In the face of the deployment of exascale supercomputing systems such as China's Tianhe-3 and Sunway Taihu Light, the United States' construction of Doudna under the Vera Rubin architecture is a key response to the global HPC dominance.
AT present, China is making rapid progress in the field of high-performance ai computing and meteorological simulation, while Europe is deploying quantum supercomputing clusters (such as Jupiter). In this context, Doudna is not only a technology implementation project, but also a "systematic response" to maintain the scientific and technological advantages of the United States in strategic fields such as defense, life sciences, and energy modeling.
Ⅵ Industrial model reconstruction: in-depth collaboration between enterprises and the national scientific research system
The particularity of the Doudna project also lies in its "government + enterprise + laboratory" three-way collaborative model, which has strong replication and demonstration in the construction of future scientific and technological infrastructure
* NVIDIA: Provide platform-level chip architecture and software stack to promote the evolution of business technology to scientific research;
* Dell: Improve deployment efficiency and carbon neutrality through standardized liquid-cooled server integration and delivery solutions;
* LBNL: As a scientific research application center, it provides feedback for the adaptation of system models and algorithms, and builds a closed-loop of scenario verification;
* DOE: Provide funding, strategic planning and technical route guarantee to strengthen national scientific and technological security and long-term technological advantages.
This model is not only applicable to HPC, but can also be extended to frontier fields such as life sciences, synthetic biology, and precision climate engineering, accelerating the two-way transformation between scientific research and industry.
Ⅶ Risks and Challenges: The Road to Openness and Integration
Despite the great prospects, Doudna faces a number of real-world challenges that need to be addressed:
* Energy consumption: Even with liquid cooling technology, the annual power consumption of the Exascale system is still expected to reach 10\~15MW, and how to reduce the carbon footprint through renewable energy and dispatch strategies will become the focus of decision-making.
* Heterogeneous software ecosystem lags behind: Software stack adaptation (especially AI + quantum fusion software) is still a shortcoming, and the speed of ecosystem construction needs to keep up with the hardware performance jump.
* Insufficient matching of scientific research talents: The fusion system requires compound scientific research talents, involving the intersection of HPC, AI, and quantum algorithms, resulting in a long talent training cycle and insufficient supply.
In the future, if Doudna can establish a modular and open architecture to support the access of universities and international research institutions, it will greatly amplify its scientific research efficiency and global influence.
Ⅷ Conclusion: Doudna is more than just computing power
Doudna isn't just another supercomputer that has reached the top of its game. It represents the transformation of the scientific research paradigm from "high-speed computing" to "intelligent reasoning", and reflects the trend of infrastructure evolution to "platform + ecology".
In the complex era when the world is facing the climate crisis, energy transition, and infectious disease prevention and control, Doudna represents a more forward-looking scientific and technological response mechanism - using intelligent computing to promote scientific breakthroughs, industrial transformation, and social governance upgrading.
When it is launched as scheduled in 2026 and ranks among the top of the global TOP500 list, Doudna will not only be the name of a machine, but also a concentrated embodiment of the United States' scientific and technological strategy and industrial synergy, and may also become a key starting point for global scientific research and computing to move towards the "era of converged intelligence".
