Skip to content
AI.info

The Pulse

NVIDIA plans Vera CPU racks for scientific computing in 2027

NVIDIA's planned scientific computing rack combines 192 Vera CPUs with high-speed networking and liquid cooling. The company expects the racks to be available from hardware partners beginning in early 2027.

NVIDIA plans Vera CPU racks for scientific computing in 2027

NVIDIA plans to bring 192 Vera CPUs together in a liquid-cooled rack for scientific computing, with systems expected from hardware partners beginning in early 2027. The company says the platform is intended for work including climate modeling, molecular dynamics and AI workflows.

The November 16 announcement pairs the processors with NVIDIA Quantum-X800 InfiniBand networking and scientific software. Dell Technologies, HPE, Lenovo and Supermicro are expected to offer the racks.

A rack for simulation

Each Vera CPU has 88 custom NVIDIA Olympus cores, 176 threads and serviceable SOCAMM2 LPDDR5X memory. The memory delivers up to 1.2TB/s of bandwidth per CPU.

The planned rack combines approximately 300TB of memory with 1 petabyte of storage in a fully liquid-cooled NVIDIA MGX design. Each 1U tray integrates eight Vera CPUs, NVMe storage and management.

The system delivers more than 200TB/s of aggregate peak memory bandwidth. It reuses NVL72 rack components, including the chassis, power distribution and cooling infrastructure. NVIDIA says its SHARP In-Network Computing technology reduces the time processors spend coordinating work during complex simulations.

Benchmark claims

In NVIDIA's tests, Vera achieved 1.9x the performance and 2.4x the performance per watt of x86 systems on OpenFOAM's motorBike Large airflow simulation. Those figures apply to that named workload rather than all scientific applications.

For the open-source Weather Research and Forecasting model, the company projected 1.6x performance and 2x energy-efficiency gains over x86. Those weather-model figures are projections, while the airflow results are described as tests.

NVIDIA introduced Vera in March for workloads including agentic AI, reinforcement learning and data processing. At that time it also announced a separate Vera CPU rack integrating 256 liquid-cooled processors for AI workloads.

Climate system and software

The German Climate Computing Center, known as DKRZ, has selected Vera CPUs for its HLRE-5 supercomputer, alongside NVIDIA GB200 Grace Blackwell systems and Quantum-2 InfiniBand networking. Bull is building the system, which is expected to be available in early 2028 for climate simulation and AI research.

NVIDIA expects the system to enable global kilometer-scale climate simulations with lower energy consumption than the current system. The company also says its HPC SDK 26.11 is available now, with compilers and libraries optimized for Vera.

A public beta of the next-generation Fortran compiler is also being introduced. Built with high-performance computing centers and national laboratories, it is intended to let developers run existing Fortran programs faster on NVIDIA platforms without rewriting their code.

In March, John Cazes, director of high-performance computing at the Texas Advanced Computing Center, said the center had tested Vera on six scientific applications while preparing for deployment in its upcoming Horizon system. He described the early results as impressive.

Sources

Companies in this story

Related stories

Explore

More articles