NVIDIA Unveils 192-CPU Vera Rack for Supercomputing

The liquid-cooled rack-scale platform connects 192 Vera CPUs over NVIDIA Quantum-X800 InfiniBand networking for high-performance computing.

Updated on Nov. 16, 2026 in Data Centers

Isometric editorial illustration of a single tall compute rack with coolant tubes and cables running down its side, in muted slate, cream, and oxblood.
NVIDIA unveiled a liquid-cooled rack holding 192 Vera CPUs linked by Quantum-X800 InfiniBand, with German climate center DKRZ selected as an early customer for its HLRE-5 supercomputer. AI Illustration. Upload story photo >

NVIDIA announced its Vera CPU rack-scale platform for high-performance computing, combining its processors, Quantum-X800 InfiniBand networking and scientific software. German climate computing center DKRZ has selected Vera CPUs for its HLRE-5 supercomputer.

Why it matters

The platform packs high memory bandwidth and storage into a single rack, aiming to speed up scientific workloads such as climate modeling and airflow simulation. Its reuse of NVL72 rack components suggests a path for operators to adopt the design without rebuilding their infrastructure.

Each Vera CPU features 88 custom Olympus cores and 176 threads and delivers up to 1.2TB/s of memory bandwidth, while the full rack holds about 300TB of memory and 1 petabyte of storage. The liquid-cooled system's aggregate peak memory bandwidth exceeds 200TB/s, though the announcement did not address power draw at rack scale.

The players

NVIDIA

NVIDIA is a American semiconductor company best known for its graphics processors that now power most AI and high-performance computing workloads.

DKRZ

The German Climate Computing Center (Deutsches Klimarechenzentrum) is a German facility that provides high-performance computing services for climate science research.

The details

NVIDIA said the platform combines its processors, Quantum-X800 InfiniBand networking and scientific software, and reuses NVL72 rack components including chassis, power distribution and cooling. Its NVIDIA SHARP In-Network Computing reduces the time processors spend coordinating their work, and NVIDIA HPC SDK 26.11 is available now with compilers and libraries optimized for Vera.

Timeline

  1. November 16, 2026: NVIDIA announced the Vera CPU rack-scale platform at SC26 in Chicago.

  2. Early 2027: Vera CPU racks are expected from Dell Technologies, HPE, Lenovo and Supermicro.

  3. Early 2028: DKRZ's HLRE-5 supercomputer in Germany is expected to become available.

The Tech Race

The Vera rack reuses the chassis, power distribution and cooling of NVIDIA's NVL72 design, extending that rack-scale architecture from GPU-centric AI systems into CPU-based scientific computing. The move positions NVIDIA against x86 server platforms that have long dominated supercomputing installations.

Organizations running high-performance computing workloads may be able to evaluate Vera CPU racks from major server vendors beginning in early 2027. The reuse of NVL72 components could simplify upgrades for operators already running that design.

The takeaway

Benchmark gains like 1.9x performance and 2.4x performance per watt suggest that CPU-based systems can still compete on efficiency for scientific workloads. Buyers planning for 2027 and 2028 deployments should compare vendor results on their own applications before committing.

What happens next

The HLRE-5 supercomputer is expected to become available in early 2028, and Vera CPU racks from Dell Technologies, HPE, Lenovo and Supermicro are expected beginning in early 2027.

Further reading

For related infrastructure coverage, see Data Centers.