The Department of Energy (DOE) is planning a new supercomputer to boost research across science. The initiative shows how commercial AI and scientific discovery are linked.

The advanced system, called “Doudna,” will be located at Lawrence Berkeley National Laboratory and should be running by 2026. It is named after Nobel laureate Jennifer Doudna for her work on CRISPR gene editing.

Dell Technologies will deliver the Doudna supercomputer, which is a change in government-funded high-performance computing.

While Hewlett Packard Enterprise has been a leader in this area, Dell’s win is a new step. “A big win for Dell,” saeid Addison Snell, CEO of Intersect360 Research, to The New York Times, pointing out Dell’s limited history here.

Dell executives said that the Doudna project allowed them to move past building custom systems for each lab. Rather, they made a flexible platform for many users. “This market had shifted into some form of autopilot. What we did was disengage the autopilot,” said Paul Perez, senior vice president and technology fellow at Dell.

The Perlmutter supercomputer at the National Energy research Scientific Computing Center at Lawrence Berkeley National Laboratory.

Nvidia‘s Vera Rubin Platform at the Core

Doudna will use nvidia’s Vera Rubin platform, designed to combine scientific simulations with AI. Unlike past DOE supercomputers that used Intel or AMD processors, Doudna will have a general-purpose Arm-based CPU from Nvidia, along with the company’s Rubin AI chips for AI and simulation tasks.

“Doudna is a time machine for science – compressing years of discovery into days.”

– Jensen Huang, founder and CEO of Nvidia

The design is made for the lab’s 11,000 users, who need both precise modeling and fast AI data analysis.

Jensen Huang, founder and CEO of Nvidia, said that “Doudna is a time machine for science – compressing years of discovery into days,” and that it will let “scientists delve deeper and think bigger to seek the basic truths of the universe.”

Ten Times Faster Than Current Systems

Doudna should be over 10 times faster than the lab’s current system, making it the DOE’s best tool for training AI models and doing simulations. Jonathan Carter, associate lab director for computing sciences at Berkeley Lab, said that the system’s design was based on the changing needs of researchers, who are using AI to improve simulations in areas like geothermal energy and quantum computing.

Doudna’s design shows a change in supercomputing. Conventional systems focused on 64-bit calculations for accuracy,but AI benefits from lower-precision operations (like 16-bit or 8-bit) for faster speeds. Dion Harris, Nvidia’s head of data center product marketing, said that combining different precision levels opens new doors for research.

The supercomputer will also connect to the Energy Sciences Network, letting researchers stream data into Doudna for real-time analysis. Sudip Dosanjh, director of the National Energy Research Scientific Computing Center, said it is “designed to accelerate a broad set of scientific workflows.”