SoylentNews

upstart [soylentnews.org] writes:

████ # This file was generated bot-o-matically! Edit at your own risk. ████

for jan

Nvidia will help build 7 AI supercomputers for for DoE [theregister.com]:

The US Department of Energy is partnering with Nvidia and Oracle to build seven new AI supercomputers to accelerate scientific research and develop agentic AI for discovery. Two of these systems, located at Argonne National Laboratory, will together form the DOE's largest AI supercomputing infrastructure.

Speaking at his company's annual GPU Technology Conference, Nvidia CEO Jensen Huang announced [nvidia.com] a partnership with Oracle and Illinois-based Argonne to build the DoE's largest-ever AI supercomputer in the form of Solstice, a 100,000 Blackwell GPU system. When interconnected with another new system Nvidia and Oracle are building at Argonne, the 10,000 Blackwell GPU-strong Equinox, the interconnected supercomputers will have a combined 2,200 exaFLOPs of AI compute performance.

"Together with Oracle, we're building the Department of Energy's largest supercomputer that will serve as America's engine for discovery, giving researchers access to the most advanced AI infrastructure to drive progress across fields ranging from healthcare research to materials science," Huang said of the duo's plans at Argonne.

Solstice and Equinox won't just be serving up more compute power for scientific experiments at Argonne, though; they're also going to be part of the lab's drive "to develop agentic scientists," Nvidia noted. While not providing much in the way of details as to what that means for scientific research at the lab, Nvidia noted that the goal of introducing agentic AI to DoE scientific research was focused on "boosting R&D productivity and accelerating discovery enabled by public research dollars within a decade."

Equinox is expected to come online next year, but neither Nvidia nor the DoE gave a timeline for the giant Solstice system's arrival on the AI supercomputing scene.

Palantir and Nvidia partner to peer into your data

Along with those national laboratory supercomputing announcements, Nvidia also announced [nvidia.com] a new partnership with Palantir at GTC today that will see the latter firm integrate Nvidia accelerated computing, CUDA-X libraries and Nvidia's Nemotron AI models into its AI Platform.

The idea behind the partnership is to give Palantir systems the ability to comb through data more quickly - and inject a shot of customizable AI agents into Palantir's systems, too.

"We're creating a next-generation engine to fuel AI-specialized applications and agents that run the world's most complex industrial and operational pipelines," Huang said of the announcement.

The combined technology already has a customer in the US hardware store chain Lowe's, which is now testing the combined tech on a digital twin of its supply chain network with the end goal of AI-driven optimization.

Argonne is also planning to launch three other [nvidia.com] new Nvidia-based systems at the lab called Tara, Minerva, and Janus. Little about those systems was mentioned beyond the company saying that they would be open to researchers at other facilities in order to expand access to AI-driven supercomputing for those without a locally based machine.

Los Alamos gets in on the Nvidia AI action, too

Not to be outdone by Argonne's plan for a giant new Nvidia AI supercomputer, Los Alamos National Laboratory in New Mexico is getting a pair [lanl.gov] of new Nvidia Vera Rubin-based systems, too, even though it just stood up a big Nvidia AI system at the lab last year in the form of the Venado supercomputer [theregister.com].

• DGX Spark, Nvidia's tiniest supercomputer, tackles large models at solid speeds [theregister.com]
• Nvidia still needs Taiwan even as TSMC ramps Blackwell production in Arizona [theregister.com]
• AI agents? Yes, let's automate all sorts of things that don't actually need it [theregister.com]
• Nvidia's GB10 workstations arrive with 1 petaFLOPS of compute, 128GB of VRAM, and a $3K+ price tag [theregister.com]

Vision, one of the two new systems coming to LANL in 2027, "will build on the success" of Venado, a Los Alamos spokesperson told us, though what exactly that means wasn't explained. Vision will be used for unclassified workloads in the national security, materials and nuclear science, energy modeling, and biomedical research fields, the laboratory said.

The other new system, Mission, will similarly be AI-powered and built using Nvidia's Vera Rubin platform. Both systems are also being built by HPE using its Cray Supercomputing GX5000 platform, but Mission will be used for classified national security science workloads and will replace the Crossroads supercomputer, which just came online at LANL in 2023 [theregister.com]. Like Vision, Mission is slated to roar to life in 2027.

"The Mission and Vision systems represent a significant investment in our national security science and basic science capabilities," LANL director Thom Mason said in the lab's statement. "These systems are purpose-built for supercomputing in the AI era."

Whether anyone has bothered to solve the AI agent accuracy problem [theregister.com] before cramming them into scientific workloads is another question altogether. ®

Get ourTech Resources [theregister.com]ShareMore about

• AI
• GTC
• Hardware

More like these×More about

• AI
• GTC
• Hardware
• Nvidia
• Supercomputer

Narrower topics

• AIOps
• Cisco ACE
• DeepSeek
• Digital advertising displays
• Gemini
• Google AI
• GPT-3
• GPT-4
• Integrated Circuit
• Keyboard
• Large Language Model
• Legacy Technology
• Machine Learning
• MCubed
• Monitor
• Neural Networks
• NLP
• Omniverse
• PCIe
• Printer
• Retrieval Augmented Generation
• Star Wars
• Supercomputing
• Tensor Processing Unit
• TOPS
• Trusted Platform Module
• USB

Broader topics

• Computer
• GPU
• Self-driving Car

More about ShareMore about

• AI
• GTC
• Hardware

More like these×More about

• AI
• GTC
• Hardware
• Nvidia
• Supercomputer

Narrower topics

• AIOps
• Cisco ACE
• DeepSeek
• Digital advertising displays
• Gemini
• Google AI
• GPT-3
• GPT-4
• Integrated Circuit
• Keyboard
• Large Language Model
• Legacy Technology
• Machine Learning
• MCubed
• Monitor
• Neural Networks
• NLP
• Omniverse
• PCIe
• Printer
• Retrieval Augmented Generation
• Star Wars
• Supercomputing
• Tensor Processing Unit
• TOPS
• Trusted Platform Module
• USB

Broader topics

• Computer
• GPU
• Self-driving Car

TIP US OFF

Send us news [theregister.com]

Original Submission