New Doudna supercomputer at Berkeley lab to power AI research
A new supercomputer meant to power artificial intelligence will soon be built for the Lawrence Berkeley National Laboratory in partnership with the Department of Energy, Dell Technologies and Nvidia.
"We're going to take a giant step up in several areas in high performance computing for scientific computing. But also, artificial intelligence as well as quantum classical computing," said Nvidia CEO Jensen Huang.
Huang was joined by U.S. Energy Secretary Chris Wright who made the announcement last week at the Berkeley Lab.
"It's going to lead to tremendous advancement in science, and it's also going to play a role in national defense," said Wright. "And that is what makes it so critical that the United States lead in artificial intelligence."
The new computer will be called Doudna, named after UC Berkeley professor Jennifer Doudna, who was awarded the Nobel Prize in 2020 for her work in gene editing technology CRISPR.
"I can't wait to see what Doudna is going to calculate," she said. "For me, it really signifies the coming together of computing and biology. This is the future. This is how the next breakthroughs are going to be made."
The Doudna computer will be built by next year and will go online and become available to scientists in 2027.
Currently, the Berkeley Lab is home to the Perlmutter supercomputer, which is the 19th fastest computer in the world, and part of the National Energy Research Supercomputing Center, known as NERSC, which connects 11,000 scientists worldwide.
"So, NERSC is available for anyone who is researching a scientific problem that is related to the Department of Energy's mission," said Jonathan Carter, Associate Lab Director for Computing Sciences at Lawrence Berkeley National Laboratory.
Those missions include creating nuclear fusion simulations, biological research, climate projections, and even mapping the universe.
"So, imagine you have a really souped-up laptop or a gaming PC, and imagine you could put 10,000 of them together tightly, so that any one program that you are running could execute on all those 10,000 CPUs at the same time," explained Carter.
Once the Doudna supercomputer is built, it is expected to be at least 10 times faster than the Perlmutter computer.
Currently, the world's fastest computer is El Capitan, which is housed also in the Bay Area at the Lawrence Livermore National Laboratory.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Epoch Times
43 minutes ago
- Epoch Times
Broadcom Shares Fall Despite Better-Than-Expected Earnings
Broadcom Inc., whose shares are trading just below a new all-time high, narrowly beat Wall Street expectations on June 5 after the Silicon Valley chipmaker and AI infrastructure giant reported its fiscal second-quarter earnings following the market's close in New York. For the period ended May 4, the Palo Alto, California-based company
Forbes
an hour ago
- Forbes
SAP Insights Newsletter: Forget LLMs. SLMs Might Be What You Need.
Artificial intelligence, data mining, modern computer technologies. SLM, Small Language model, Generative AI Small vs. large: Not every company needs a large language model. Small language models (SLMs) are great tools for specific use cases. They're also quicker, cheaper, and easier to add to your tech portfolio. We discuss why some companies are choosing SLMs and how to figure out if an SLM is right for your organization. Flex vs. fixed: B2B is trending toward B2C pricing strategies, specifically flex pricing. Many businesses are finding that flex pricing has many advantages over fixed pricing. With global events, trade, tariffs, and the like, being able to adjust pricing is how companies protect their revenue. Productivity vs. GenAI: World Productivity Day is meant to inspire us to reach new levels of GTD (getting things done). Yet, where does GenAI fit in? The productivity promises of GenAI tools are, shall we say, maybe a little unrealistic (at least right now). We explain the challenges and where GenAI tools can make the most impact right now. Research that hasn't reached your inbox: We discover the importance of colleague support; how labor vs. luck is viewed by employees; and a promising new approach to mental health.
Forbes
an hour ago
- Forbes
How AI Can Decode The Hidden Stories In Immigration Applications
Raghu Para is a tech exec with over 15 years of progressive experience in software, artificial intelligence and machine learning. getty Picture this: You've spent years gathering documents, filling out forms and waiting for your immigration decision. Meanwhile, the officer reviewing your case is buried under a mountain of paperwork, armed with the patience of a kindergarten teacher and the attention span of a detective running on espresso. This is the modern immigration system—a finely tuned cocktail of bureaucracy, backlogs and burnout. Governments want meticulous vetting. Applicants wait so long they could've binged Living Undocumented on Netflix. And the kicker? Much of the work is mind-numbingly repetitive. Officers aren't just reviewing facts—they're decoding intent. Is this a legit work visa? A bona fide asylum claim? They play legal detective, scan for red flags and occasionally channel TSA energy—unpacking a grandma's suitcase only to find a single, compliant three-ounce shampoo bottle. Could AI help? Sure. But the question is—can it understand human intent without making a mess? Let's address the customs officer in the room: AI in immigration is controversial. On the bright side, processing times are expedited, costs can be reduced and the risk of human error deciding anyone's fate can be mitigated. But let's not hand it a rubber stamp just yet. AI bias is real. It can reject perfectly good applications like it's giving out Halloween candy—and worse, hallucinate fake laws like the "Deportation Reform Act of 2065." That's not a typo—it's fiction. So what's the answer? Let tech sit at the desk—but humans still hold the stamp. As an AI researcher who's navigated the anxiety-inducing immigration process myself, I can tell you the challenge isn't building smart algorithms—it's building guardrails that stop them from going off-script. When designed carefully, AI can be the ultimate sidekick—bringing technical muscle and just enough empathy to keep things human. Today's systems aren't the clunky chatbots that used to ask, "Did you mean refugee or retirement visa?" before crash-landing on a 404 page from the Bush administration. Modern systems combine machine precision with human oversight, making a huge difference. Here's how: One of the biggest delays in immigration comes from verifying intent. AI now uses natural language processing to read between the lines, flag inconsistencies and detect fraud. Take a framework like Agent-Driven Semantic Analysis & Intent Detection (ADS-ID)—a multi-agent model I helped design: • The "document detective" deciphers even the messiest handwriting (think doctor's prescription, but worse). • The "legal scholar" cross-references case law better than your cousin with a law degree and zero follow-through. • The "consistency checker" spots contradictions like "You were working in Canada while attending school in Mexico?" Okay, time traveler. Other approaches use deep learning trained on millions of past cases or hybrid models combining logic rules and machine learning. But the solutions always keep humans in the loop to interpret AI's findings. And yes, there are challenges. • The Black Box Problem: If an AI makes a decision, it needs a "Kindergarten Explanation Layer"—something even a five-year-old (or a policy analyst) can understand. • Biased Training Data: Immigration decisions are rooted in decades of judgment calls—many flawed. We need AI that can recognize, adapt to and correct for that. • Constant Policy Change: Immigration rules can change faster than the promises of a politician. AI needs regular policy memos just to keep up. Some fraud is obvious, but some fraud is sneaky. AI can help by analyzing digital footprints for inconsistencies and flagging suspicious patterns in application histories. The U.K.'s Whitehall system, for instance, used AI to detect sham marriages, though critics noted it sometimes flagged real couples, too. Embassies are starting to use AI to estimate wait times based on application type, country of origin and historical data. It's not flawless, but it beats refreshing your status page 37 times a day. Officers often deal with documents in rare dialects, bad translations or messy handwriting. Systems should evolve to support the human dynamic, ensuring officers operate like collaborators who use AI as a sounding board—not make them "overrulers" who distrust any algorithmic suggestion or rubber-stampers who approve whatever AI says. AI in immigration isn't perfect. It can hallucinate laws, mishandle sensitive data that deserves Fort Knox-level security or reject a case with a digital shrug: "Too complicated, goodbye." That's why we need transparency, human oversight—and maybe a big red "Don't Panic" button. But if we build with care, the future is promising. We'll see AI that analyzes video (Photoshop weddings won't cut it), uses quantum computing (finally faster than a clerk on dial-up) and sends real-time updates that don't leave you in "Pending" purgatory. Immigration is—and always will be—a human business. But with AI as a savvy, respectful assistant, officers can focus on what really matters: serving people, not pushing paper. Ultimately, the goal is to ensure it's faster, fairer and insightful so families can reunite while processes flow. Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
