In brief
- The NVIDIA Vera platform combines 192 custom-designed CPUs with a Quantum-X800 InfiniBand network.
- According to the company, the system offers more than 200 TB/s of peak memory bandwidth and approximately 300 TB of memory in total.
- NVIDIA HPC SDK 26.11 is now available; the new Fortran compiler is offered as a public beta.
- Vera CPU racks are expected to be available through Dell Technologies, HPE, Lenovo and Supermicro starting in early 2027.
NVIDIA announced its Vera CPU rack-scale platform for high-performance computing (HPC) on 16 November 2026. According to the company, the system targets compute-intensive scientific workloads such as climate modeling and molecular dynamics. The platform combines 192 Vera CPUs with the NVIDIA Quantum-X800 InfiniBand network and is based on a fully liquid-cooled NVIDIA MGX design.
What hardware is in the Vera rack?
Each Vera CPU features 88 custom-designed NVIDIA Olympus cores and 176 threads. The processors use LPDDR5X memory, which NVIDIA says delivers up to 1.2 TB/s of bandwidth per CPU, in serviceable SOCAMM2 modules. NVIDIA says the rack’s total peak memory bandwidth exceeds 200 TB/s.
The 192-processor system has approximately 300 TB of memory and 1 petabyte of storage. Each 1U tray houses eight Vera CPUs, NVMe storage and management components. According to NVIDIA, the platform reuses components from NVL72 racks—including the chassis, power distribution and cooling infrastructure—to simplify installation and servicing. NVIDIA SHARP In-Network Computing is also integrated into the network to reduce the time processors spend on coordination.
What do NVIDIA’s performance figures show?
In NVIDIA’s tests, Vera delivered 1.9x the performance and 2.4x the performance per watt of x86 systems in OpenFOAM’s motorBike Large airflow simulation. The company also says the platform is twice as energy-efficient as x86 CPUs. These results are from the manufacturer’s own tests; the source points to NVIDIA’s performance document for details of the test configurations.
Comparisons were also shared for weather forecasting and climate research. NVIDIA projects 1.6x the performance and twice the energy efficiency of x86 in the open-source Weather Research and Forecasting model. Germany’s DKRZ climate computing center selected Vera CPUs for its HLRE-5 supercomputer. The system, expected to be built by Bull and enter service in early 2028, will also include NVIDIA GB200 Grace Blackwell systems and a Quantum-2 InfiniBand network.
Software and availability timeline
NVIDIA HPC SDK 26.11 is now available with compilers and libraries optimized for Vera. NVIDIA also announced a public beta of its next-generation Fortran compiler, developed with research centers and national laboratories. According to the company, the compiler is designed to help existing Fortran programs run faster on NVIDIA platforms without rewriting their code.
Vera CPU racks are expected to be available from Dell Technologies, HPE, Lenovo and Supermicro starting in early 2027. The source does not provide pricing for the platform.
Impact on architecture and visualization workflows
Vera was not announced as a product for desktop visualization or architectural modeling; its focus is scientific HPC workloads. It should therefore not be viewed as a direct workstation upgrade for architecture and interior design firms. However, the platform may interest organizations that use research and computation to inform design decisions, including those working on climate modeling and large-scale simulations.
For architectural visualization teams, the key questions to watch are whether Vera-specific software and libraries are compatible with the applications they use, what the infrastructure requires for installation, and what pricing will be announced. The source provides no information about compatibility with architectural software or performance gains in those applications.
Sources
1 sourceSource texts are not republished; short quotes are marked, everything else is our own summary and commentary.
There is no information indicating that Vera is currently being presented as a direct modeling or rendering solution for architecture firms. The announcement’s main value is its integrated approach to CPU, memory and networking infrastructure for compute-intensive scientific workloads. The platform’s purpose appears more relevant to research teams working in areas such as climate simulation.
For firms and visualization teams in Turkey, pricing, availability and compatibility with existing software should be clarified before making a purchase decision. The stated supply timeline of early 2027 is not, by itself, a reason to invest. NVIDIA’s performance results are based on the company’s own tests; independent measurements on real workloads and the total infrastructure requirements will matter.
Frequently asked questions
When are NVIDIA Vera CPU racks expected to be available?
They are expected to be available starting in early 2027 through Dell Technologies, HPE, Lenovo and Supermicro. The source does not disclose pricing.
How many CPUs are in the NVIDIA Vera platform?
A Vera rack contains 192 CPUs. Each CPU has 88 NVIDIA Olympus cores and 176 threads.
Can NVIDIA’s new Fortran compiler work with existing code?
NVIDIA says its public beta compiler is designed to run existing Fortran programs on NVIDIA platforms without requiring the code to be rewritten.



