NVIDIA has announced that its first standalone CPU, Vera, will power three new supercomputers, Mission, Vision and Veritas, set to be deployed at Los Alamos National Laboratory (LANL) in the United States[1]. All three are being built together with HPE as computing platforms tailored for scientific research, with operations expected to begin in 2027. In tests at the laboratory, the Vera CPU delivered up to 7x the performance of the x86-based CPUs in the lab's current supercomputer, opening a new option for the computational science community.
Three Systems for Science, Built on the Vera Rubin Platform
The three supercomputers combine HPE's Cray Supercomputing GX5000 architecture with NVIDIA's Vera Rubin platform[1]. Vera Rubin brings together the Vera CPU, the Rubin GPU, and the high-speed Quantum-X800 InfiniBand networking, aiming to accelerate computation and communication as a single unit.
Mission, the centerpiece, will include Vera Rubin GPU nodes plus 2,300 standalone Vera CPUs without GPUs[1]. Mission is scheduled to come online in 2027, serving as the fifth Advanced Technology System in the Advanced Simulation and Computing program of the National Nuclear Security Administration (NNSA), replacing the current Crossroads system to handle classified national security workloads[1].
Veritas will carry roughly 1,150 standalone Vera CPUs to complement the Vera Rubin nodes[1]. It supports the laboratory's own research and development program and is tasked with accelerating agentic AI for science. It also serves as a testbed for new technologies ahead of building out larger systems. The third system, Vision, is likewise slated for 2027 and will be used as a computing resource open to a wide range of fundamental science, including materials science, nuclear science, energy modeling, biomedical research and AI[1]. HPE itself has confirmed that it is building the Mission and Vision systems[2].
Up to 7x Over x86: What the Vera CPU Can Do
What drove this adoption was the Vera CPU performance the laboratory measured in real scientific computing. In LANL's testing, on workloads for URSA, an AI framework for scientific research, the Vera CPU delivered 7x the performance of the x86-based CPUs used in the current Crossroads system[1].
In early testing with Branson, an open source Monte Carlo heat transfer simulation tool, Vera outperformed the x86-based CPUs in Crossroads by more than 3x[1]. These results are underpinned by design choices such as NVIDIA's in-house Olympus core, low-power LPDDR5 memory, and a fast on-chip fabric that links the chip internally at high speed.
According to NVIDIA, a single Vera CPU offers more than 3x the performance of a single-socket x86 CPU, with more than 4x the memory per core and 6x the memory per node[1]. Beyond raw compute, it is a configuration suited to scientific computing that processes work while holding large volumes of data. These systems were built through codesign, in which hardware architects, system software developers, domain scientists, computer scientists and applied mathematicians refine the design together, so that the systems are optimized around real research workloads rather than abstract benchmarks alone[1].
Agentic AI and URSA: Driving Science Autonomously
What the three supercomputers have their eye on is using agentic AI to advance scientific research itself, autonomously. Researchers will be able to hand an AI agent a chain of work, forming hypotheses, choosing the tools to use, running simulations, analyzing the results that come out, and refining the next step[1].
At the center of this is URSA (Universal Research and Scientific Agent), which LANL is developing in the open. It is a modular, feedback-driven AI framework that supports everything from generating hypotheses to planning experiments, running simulations and analyzing results. It already runs on the existing Venado system, with plans to run on Mission and Vision going forward[1].
This effort did not begin out of nowhere. LANL and NVIDIA have collaborated on CPUs for more than a decade, from Grace to Vera, closely aligning their designs for the laboratory's simulation work[1]. The three new systems sit on the trajectory of Venado, the HPE Cray EX system installed in 2024 with NVIDIA's GH200 Grace Hopper Superchips and Grace CPU Superchips[1].
Summary
NVIDIA's standalone Vera CPU has been chosen for Los Alamos National Laboratory's new supercomputers, Mission, Vision and Veritas. Built together with HPE, they are expected to come online in 2027. In testing they showed up to 7x the performance of x86-based CPUs, positioning them as a foundation for agentic AI that advances science autonomously and for large-scale simulation. NVIDIA, which has built a strong presence with GPUs, has now extended its footprint into the important arena of scientific computing on the CPU side as well.
Source:https://blogs.nvidia.com/blog/nvidia-vera-cpu-los-alamos-national-laboratory/
