Argonne National Laboratory has expanded its Argonne Leadership Computing Facility Inference Service with a new system called Tara. Built with Hewlett Packard Enterprise on Cray EX4000 architecture, Tara includes 672 nodes and 2,688 NVIDIA GH200 Grace Hopper Superchips, delivering roughly 100 times more GPU memory than the previous platform. The lab also added Minerva with 64 NVIDIA B200 GPUs and expanded its Metis platform with SambaNova chips. The upgrades support more than 50 advanced models in a secure on-premises environment and enable agentic workflows that plan experiments, write code, and steer lab instruments. Early uses include chemistry automation and autonomous beamline experiments under the Department of Energy’s Genesis Mission. Researchers access the shared resource through APIs or chat interfaces, reducing the barrier for individual teams to apply large AI models to science, energy, and national security problems.
Argonne National Laboratory has expanded its Argonne Leadership Computing Facility Inference Service with a new system called Tara. Built with Hewlett Packard Enterprise on Cray EX4000 architecture, Tara includes 672 nodes and 2,688 NVIDIA GH200 Grace Hopper Superchips, delivering roughly 100 times more GPU memory than the previous platform. The lab also added Minerva with 64 NVIDIA B200 GPUs and expanded its Metis platform with SambaNova chips. The upgrades support more than 50 advanced models in a secure on-premises environment and enable agentic workflows that plan experiments, write code, and steer lab instruments. Early uses include chemistry automation and autonomous beamline experiments under the Department of Energy’s Genesis Mission. Researchers access the shared resource through APIs or chat interfaces, reducing the barrier for individual teams to apply large AI models to science, energy, and national security problems.