LANL selects NVIDIA Vera CPUs for three supercomputers, claiming 7x performance on agentic AI workloads over x86. Systems Mission, Vision, Veritas deploy by 2027.
Los Alamos National Laboratory selected NVIDIA Vera CPUs for three new supercomputers, claiming a 7x performance gain on agentic AI workloads over x86. The systems—Mission, Vision, and Veritas—will deploy the HPE Cray GX5000 architecture with NVIDIA Rubin GPUs and Vera CPUs by 2027.
Key facts
- Vera CPU delivers 7x higher performance on URSA agentic AI workloads.
- Mission will include 2,300 standalone Vera CPUs using HPE Cray GX240 blades.
- Veritas features ~1,150 standalone Vera CPUs alongside Vera Rubin nodes.
- Vera outperforms x86 by over 3x on Branson Monte Carlo simulation.
- Mission and Vision expected operational in 2027.
- Vera provides 4x memory per core and 6x memory per node vs x86.
Key Takeaways
- LANL selects NVIDIA Vera CPUs for three supercomputers, claiming 7x performance on agentic AI workloads over x86.
- Systems Mission, Vision, Veritas deploy by 2027.
Vera CPUs Deliver 7x on Agentic AI Benchmarks
Los Alamos National Laboratory (LANL) announced plans to build three supercomputers—Mission, Vision, and Veritas—using the NVIDIA Vera CPU platform. The systems will replace the Crossroads x86 supercomputer and target agentic AI for scientific discovery, including materials science, nuclear modeling, and biomedical research.
In early testing, the Vera CPU delivered 7x higher performance on URSA (Universal Research and Scientific Agent) workloads compared to the CPUs in Crossroads. URSA is a modular, feedback-driven AI framework designed to autonomously form hypotheses, choose tools, launch simulations, analyze outputs, and refine subsequent steps. LANL also reported that Vera outperformed x86 CPUs by over 3x on Branson, an open-source Monte Carlo heat transfer simulation tool.
Architecture and Deployment Details
Mission will include NVIDIA Vera Rubin GPU nodes and 2,300 standalone Vera CPUs using the HPE Cray GX240 blade. Veritas will feature approximately 1,150 standalone Vera CPUs alongside Vera Rubin nodes, serving the Laboratory Directed Research and Development program. Vision will complement both as a resource for fundamental science and AI method development.
The performance gains stem from Vera's custom Olympus core, LPDDR5 memory, and fast on-chip fabric. A single Vera CPU provides more than 4x the memory per core and 6x the memory per node compared to a single-socket x86 CPU, according to NVIDIA. All systems were codesigned by hardware architects, system software developers, domain scientists, computer scientists, and applied mathematicians to ensure workloads—not abstract benchmarks—shaped the design.
Historical Context and Competitive Positioning
Mission, expected operational in 2027, will be the fifth Advanced Technology System in the National Nuclear Security Administration's Advanced Simulation and Computing program. It will handle classified national security workloads. The announcement extends a decade-long collaboration between LANL, HPE, and NVIDIA, following the Venado system that earlier tested NVIDIA Grace Hopper superchips.
This marks a direct challenge to Intel and AMD in the HPC CPU market, where x86 has dominated for decades. NVIDIA's Vera CPU, part of the broader Vera Rubin platform, is architected specifically for AI and simulation workloads rather than general-purpose computing. The 7x performance claim on agentic AI tasks suggests that NVIDIA is betting on specialized compute to displace traditional CPUs in scientific computing, mirroring its strategy in the GPU market.
What to Watch
Watch for independent benchmark results from LANL after Mission and Vision go operational in 2027, particularly URSA performance comparisons against AMD's upcoming Turin CPUs and Intel's Granite Rapids. Also monitor whether other national labs follow suit with Vera CPU deployments, and whether NVIDIA releases Vera CPU performance data on standard HPC benchmarks like HPL and HPCG.
Source: blogs.nvidia.com
Originally published on gentic.news
Top comments (0)