NVIDIA expands Vera CPU platform for high-performance computing

NVIDIA announced the NVIDIA Vera CPU rack-scale platform for high-performance computing on 16 November 2026, combining a custom data center CPU with NVIDIA Quantum-X800 InfiniBand networking. The system delivers 1.9x the performance and 2.4x the performance per watt over x86 systems for certain simulations, and it matters to any organization scaling non-GPU scientific workloads. If you already use standard x86 systems for scientific modeling, none of this changes your next purchase until you evaluate the specific efficiency gains for your model.

Is this for you?

If you run non-GPU scientific workloads, the Vera platform offers 1.9x the performance and 2.4x the performance per watt over x86 systems.

  • If you use standard x86 systems for scientific modeling, the 1.9x performance and 2.4x energy-efficiency gains are the key metrics.
  • If you use Fortran for scientific computing, the new public beta compiler allows you to run existing code without rewrites.
  • If you manage data centers, the 192-CPU configuration requires a liquid-cooled MGX design to handle thermal output.
  • Skip this if you build gaming PCs as the Vera platform is a rack-scale data center solution.
  • Skip this if you manage standard enterprise IT as this is for unaccelerated scientific workloads rather than office applications.

What you can do now

  1. If you use Fortran, download the public beta Fortran compiler to test your current library's performance.
  2. If you are an infrastructure manager, coordinate with Dell, HPE, Lenovo, or Supermicro to secure capacity for early 2027.
  3. If you are an HPC architect, verify that your facility's liquid cooling standard supports the NVIDIA Vera CPU rack requirements.

Key figures

NVIDIA Olympus cores
88 cores per CPU
Threads per CPU
176 threads
Memory bandwidth
1.2TB/s per CPU, 16 November 2026
Rack memory capacity
300TB, 16 November 2026
Storage capacity
1 petabyte
Availability
Early 2027

NVIDIA launches its first custom data center CPU with the Vera platform

The NVIDIA Vera CPU is a fully custom data center CPU designed to drive extreme efficiency and throughput for state-of-the-art scientific modeling, simulation, and agentic science—science driven by autonomous AI agents. This platform addresses bottlenecks in scientific simulations where CPUs must keep large datasets moving as applications scale. NVIDIA says the Vera platform brings together custom CPU cores, high-bandwidth memory, and InfiniBand networking to solve these issues for researchers.

NVIDIA's move to custom silicon signals a shift toward unaccelerated

NVIDIA’s move to custom silicon signals a shift toward unaccelerated, next-generation scientific workloads—computational tasks that run on CPUs rather than GPUs. (unaccelerated: computational tasks that run on central processing units rather than graphics processing units.) By designing the Olympus cores specifically for high-concurrency scientific workloads, NVIDIA targets memory-bound tasks that standard x86 architectures struggle to process efficiently. This transition allows the company to capture the high-performance computing (HPC) market for scientific modeling where GPU acceleration is not the primary requirement.

Core hardware specifications of the Vera CPU

Each Vera CPU features 88 custom NVIDIA Olympus cores and 176 threads; the 2:1 thread-to-core ratio indicates a design optimized for high concurrency in multi-threaded scientific workloads. The processor uses SOCAMM2 LPDDR5X—high-speed memory sticks—to deliver up to 1.2TB/s of memory bandwidth per CPU. The use of SOCAMM2 provides a compact form factor that supports the high memory throughput required for data-intensive science.

The rack-scale architecture combines 192 Vera CPUs with approximately 300TB of memory and 1

The rack-scale architecture combines 192 Vera CPUs with approximately 300TB of memory and 1 petabyte of storage. The system is built in a fully liquid-cooled NVIDIA MGX design; the high density of the 192-CPU configuration necessitates liquid cooling to manage the thermal output of the rack. This design allows for extreme density while maintaining the thermal stability required for continuous HPC operations.

Comparative performance gains over x86 architecture

NVIDIA performed tests showing the Vera CPU delivered 1.9x the performance and 2.4x the performance per watt over x86 systems on OpenFOAM’s motorBike Large airflow simulation. In these tests, the Vera platform demonstrated its ability to handle complex fluid dynamics more efficiently than standard processors.

For climate research

For climate research, NVIDIA benchmarked the open source Weather Research and Forecasting (WRF) model on Vera CPUs. The platform is projected to deliver 1.6x the performance and 2x the energy-efficiency gains compared with x86 systems. These gains allow for global kilometer-scale climate simulations with lower power consumption than current systems.

Software ecosystem and the new Fortran compiler

NVIDIA HPC SDK 26.11 is now available to provide compilers and libraries optimized for the Vera platform. This software layer is essential for researchers to translate high-level scientific requirements into optimized machine code.

NVIDIA is also introducing a public beta of a next-generation Fortran compiler

NVIDIA is also introducing a public beta of a next-generation Fortran compiler, built in collaboration with leading HPC centers and national labs. This compiler allows developers to run existing Fortran code—the standard language for scientific computing—on NVIDIA platforms without needing to rewrite their code. By preserving legacy libraries, NVIDIA removes a major barrier to adopting its custom silicon.

Vendor availability and deployment timelines

Vera CPU racks are expected to be available from Dell Technologies, HPE, Lenovo, and Supermicro beginning in early 2027. This timeline provides a clear window for organizations to plan their infrastructure upgrades for the next HPC cycle.

The source did not state the specific cost of the Vera CPU racks or the various configurations available. Furthermore, the source did not specify how these hardware specifications differ from previous generations of NVIDIA data center processors.

HPC System Architects verifying liquid cooling

HPC System Architects verifying liquid cooling are for those who need to verify that the facility’s liquid cooling standard supports the NVIDIA Vera CPU rack requirements. Because the 192-CPU configuration requires a liquid-cooled MGX design, architects must ensure the infrastructure can handle the thermal output. This holds while the system remains a liquid-cooled solution; if the facility is restricted to air cooling, the Vera platform cannot be deployed.

Climate Research Scientists watching WRF

Climate Research Scientists watching WRF are for those who need to watch for the official WRF benchmark results on the NVIDIA Vera CPU platform. Since the Vera platform is projected to deliver 2x energy-efficiency gains over x86 for climate modeling, these results will dictate the viability of large-scale climate simulations. The decision to adopt the platform rests on these specific efficiency gains; if production benchmarks fail to reach these targets, the choice favors existing x86 systems.

Scientific Software Developers using Fortran

Scientific Software Developers using Fortran are for those who need to download the public beta Fortran compiler to test existing code on the NVIDIA Vera CPU. Because the compiler is designed to run existing programs without rewrites, developers can immediately begin evaluating their current library’s performance on the new hardware. The decision to move to Vera hinges on the compiler’s ability to maintain accuracy while increasing speed.

Procurement managers for 2027

Infrastructure Procurement Managers for 2027 are for those who need to schedule the hardware procurement for the NVIDIA Vera CPU rack for the early 2027 cycle. Because the platform will be available through Dell, HPE, Lenovo, and Supermicro, managers should coordinate with these vendors to secure capacity. If the 2027 availability window shifts, the procurement plan for the HLRE-5 supercomputer or similar projects may need to be delayed.

Gaming PC Builders and standard IT

Gaming PC Builders and standard IT can ignore this and should instead look at NVIDIA GeForce GPU news. The Vera platform is a rack-scale data center solution, not a consumer desktop CPU. Standard Enterprise IT can also ignore this and should instead look at NVIDIA Grace or general server CPU news, as Vera is specialized for unaccelerated scientific workloads rather than general office applications.

A climate research scenario for Mateo

Mateo runs a national research institute and needs to run complex climate simulations using the WRF model. He currently uses an x86-based supercomputer but is concerned about the increasing energy costs and limited performance of the current system. Mateo evaluates the NVIDIA Vera CPU and notes the projected 1.6x performance gain and 2x energy-efficiency gain for his specific workload.

Mateo decides to select the Vera platform for his next cluster, knowing it will handle his simulations faster while staying within his energy budget. However, if Mateo’s primary need was to run standard web applications rather than scientific simulations, he would ignore the Vera platform and stick with a standard server CPU.

Applications for climate and molecular research

Use Case Who it serves
Global kilometer-scale climate simulations Climate scientists at centers like DKRZ
Simulating molecular dynamics Molecular biologists or chemists
Migrating legacy research code Research scientists with legacy Fortran libraries
Liquid-cooled HPC rack deployment Data center architects

Replacing x86 systems with Vera

NVIDIA Vera CPU replaces standard x86 systems for certain scientific workloads. NVIDIA says the platform delivers 1.9x the performance and 2.4x the performance per watt on OpenFOAM simulations. This shift allows researchers to move away from general-purpose processors that struggle with memory-bound tasks, moving instead to custom silicon that provides higher throughput and better energy efficiency for scientific modeling.

NVIDIA Vera CPU: what to check, compare and watch

x86 system
The NVIDIA Vera CPU is compared against the performance of standard x86 systems in simulations. Performance: 1.9x. HPC System Architect: compare the performance of the NVIDIA Vera CPU against the current x86 system used in the data center.
Power distribution unit
The NVIDIA Vera CPU rack reuses existing power distribution components for simplified deployment. HPC System Architect: confirm the power distribution unit compatibility for the NVIDIA Vera CPU rack deployment.
NVIDIA GB200 Grace Blackwell
The DKRZ chose NVIDIA Vera CPUs alongside NVIDIA GB200 Grace Blackwell systems for their supercomputer. Climate Research Scientist: evaluate the NVIDIA Vera CPU against the NVIDIA GB200 Grace Blackwell for the climate simulation project.
Fortran standard
The NVIDIA Vera CPU compiler is designed to support the standard language used in scientific computing. Scientific Software Developer: check the compiler documentation to ensure the Fortran standard matches the project requirements.

NVIDIA Vera CPU: questions the original report leaves open

What are the key specifications, and how do they differ from the previous generation?
The original report did not say how the specifications differ from the previous generation.
How much does it cost, and in which configurations?
The original report did not say how much it costs or the configurations.
Who should upgrade, and from what?
The original report did not say who should upgrade or from what.
How does NVIDIA Vera CPU compare to NVIDIA GB200 Grace Blackwell?
The original report does not say.

Original reporting: NVIDIA Expands Vera CPU Platform to Accelerate Scientific Discovery (NVIDIA), on 16 November 2026.

More on NVIDIA Vera CPU

ExtremeSpec is reader-supported: some links on this page are affiliate links, and as an Amazon Associate ExtremeSpec earns from qualifying purchases at no extra cost to you. Learn more.

Similar Posts