NVIDIA is taking its Vera processor deeper into scientific computing, with a rack-scale system announced on November 16, 2026 at SC26 in Chicago. The target includes climate and molecular simulations.
What the new rack brings together
The company says its liquid-cooled design connects 192 Vera CPUs through Quantum-X800 InfiniBand, with approximately 300 terabytes of memory and 1 petabyte of storage. Dell Technologies, HPE, Lenovo and Supermicro are expected to offer the racks from early 2027. This is a planned availability window, not a completed rollout.
NVIDIA also announced HPC SDK 26.11 and a public beta of a new Fortran compiler. Software matters alongside hardware: existing scientific programs need tools that can use a different processor efficiently.
A climate customer has its own timetable
The German Climate Computing Center, DKRZ, provides a concrete example of the research demand. In its October 26 announcement, the center confirmed a €40 million contract with Bull for HLRE-5, including five years of support. It expects the machine to enter service in early 2028.
DKRZ’s planned system is not identical to NVIDIA’s newly announced rack configuration. It combines a Vera CPU partition with a GB200 Grace Blackwell GPU partition and Quantum-2 InfiniBand networking. According to the center, the GPU side will handle most simulation and AI work, while the CPU partition introduces Arm architecture into its core system.

The center expects to run two global climate simulations at 5-kilometer resolution simultaneously. Its projected reduction of around 75% in energy per simulated year is relative to its current Levante system. That is a project-specific target, not a universal saving for every Vera installation.
Keep the benchmark claims in context
NVIDIA reports 1.9 times the performance and 2.4 times the performance per watt of x86 systems in its OpenFOAM airflow test. These are supplier results, not an independent guarantee for every scientific program.
- Commercial availability and a research center’s commissioning date are separate milestones.
- Energy per completed simulation differs from total electricity consumption.
- Results depend on the application, comparison system and configuration.
That last distinction also runs through our earlier report on AI research and its energy trade-offs. A useful tool and a smaller environmental footprint are related questions, but proving one does not automatically prove the other.
For AsiaNewsIran, the distinction is between announced specifications and demonstrated outcomes. Which matters most for your research: faster results, lower energy use or easier software migration?