NVIDIA Expands Vera CPU Platform to Accelerate Scientific Discovery

Glass Epoch DeskOriginal digestSource: NVIDIA Newsroom

AI-assisted digest, checked against the original source before publishing. How we work

NVIDIA's first fully custom data center CPU arrives as a rack-scale HPC platform, paired with Quantum-X800 InfiniBand and shipping through Dell, HPE, Lenovo and Supermicro.
NVIDIA Expands Vera CPU Platform to Accelerate Scientific Discovery
Photo: Brecht Corbeel · free stock license

NVIDIA today unveiled the Vera CPU rack-scale platform for high-performance computing, pairing its first fully custom data center CPU with Quantum-X800 InfiniBand networking and scientific software aimed at weather, climate and molecular simulations. The full system will be available from Dell Technologies, HPE, Lenovo and Supermicro.

The rack ships 192 Vera CPUs interconnected by Quantum-X800 InfiniBand, targeting climate modeling, molecular dynamics and other largely unaccelerated scientific workloads. Liquid-cooled in an NVIDIA MGX design, it delivers more than 200TB/s of aggregate peak memory bandwidth, roughly 300TB of memory and 1 petabyte of storage, with each 1U tray integrating eight Vera CPUs plus NVMe storage and management.

Each Vera CPU carries 88 custom NVIDIA Olympus cores, 176 threads and compact, serviceable SOCAMM2 LPDDR5X memory providing up to 1.2TB/s of bandwidth for data-intensive applications. The platform reuses NVL72 rack components — chassis, power distribution and cooling — to simplify deployment and service, while NVIDIA SHARP In-Network Computing cuts the time processors spend coordinating work on complex simulations.

In NVIDIA's own testing, Vera posted 1.9x the performance and 2.4x the performance per watt of x86 systems on OpenFOAM's motorBike Large airflow simulation, underpinning the company's claim of 2x greater energy efficiency than x86 CPUs for scientific computing. "Scientific simulations need CPUs that can keep large datasets moving and sustain performance as applications scale," said Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA, adding that custom cores, high-bandwidth memory and InfiniBand address those bottlenecks as compilers and libraries get optimized for Vera.

Watch for independent validation beyond NVIDIA's benchmark: whether the 1.9x and 2.4x gains hold across weather, climate and molecular dynamics codes rather than a single airflow case, how the platform prices against x86 HPC racks, and how quickly compilers and libraries reach maturity for Vera's custom cores. Vendor availability dates and memory configuration options were not detailed in the announcement.

Key facts

  • • 192 Vera CPUs rack with Quantum-X800 InfiniBand, 200TB/s peak bandwidth.
  • • Each Vera CPU: 88 Olympus cores, 176 threads, 1.2TB/s memory bandwidth.
  • • Rack holds ~300TB memory and 1PB storage in liquid-cooled MGX design.
  • • NVIDIA tests: 1.9x performance, 2.4x perf/watt vs x86 on OpenFOAM.
  • • Shipping via Dell, HPE, Lenovo and Supermicro.

Why it matters

Builders targeting HPC workloads get a CPU-plus-network rack that reuses NVL72 infrastructure, cutting deployment and service overhead for labs and cloud operators. The benchmark claims matter mainly if third-party tests confirm them across varied scientific codes; until then, treat the 2x efficiency figure as vendor-measured. Expect software readiness — compilers and libraries optimized for Vera

Spotted an error? Send a correction.

Get the day's AI news in one email

The Daily Epoch · 60-second read · unsubscribe anytime

Related