· Product launches· 日本語で読む
NVIDIA Announces Vera CPU Rack-Scale Platform for Scientific Computing
NVIDIA announced the Vera CPU rack-scale platform for scientific computing on November 16, 2026. It combines the company’s first fully custom-designed data center CPU with HPC networking. The platform is aimed at scientific modeling, simulation and agentic science.
Key facts
- NVIDIA announced the “NVIDIA Vera CPU rack-scale platform” for high-performance computing on November 16, 2026.
- The platform combines the company’s first fully custom-designed data center CPU with HPC networking.
- Scientific modeling, simulation and agentic science are among its target applications.
What happened
NVIDIA announced the “NVIDIA Vera CPU rack-scale platform” for high-performance computing on November 16, 2026. It combines the company’s first fully custom-designed data center CPU with HPC networking and is positioned as a platform to support scientific modeling, simulation and agentic science.
Why it matters
By presenting a rack-scale platform that combines CPUs and HPC networking for scientific modeling, simulation and agentic science, NVIDIA has made the network-inclusive system configuration relevant when evaluating these workloads, not just the CPU alone. The announcement materials include no performance figures or system specifications, so the configuration and evaluation criteria to be disclosed later will be needed to assess how it differs from conventional HPC platforms.
Analysis
The central point of the announcement is that NVIDIA presented CPUs and HPC networking as a single rack-scale platform for scientific computing. The company described the CPU as its “first fully custom-designed data center CPU” and listed scientific modeling, simulation and agentic science as target applications. The announcement materials, however, outline the direction of the configuration and its intended uses but do not provide CPU specifications or performance figures, the number of units per rack or the networking method. As a result, the announcement does not make it possible to determine how the platform’s role will compare with systems that include GPUs or what performance it will deliver in actual scientific computing workloads.
Related articles from this newspaper cover NVIDIA’s computing platforms from different angles. NVIDIA announces new GPU “Rubin HPC” for science and engineering covers GPUs for science and engineering, while NVIDIA marks CUDA’s 20th anniversary at SC26, looks back at its contributions to scientific computing and AI covers CUDA. Like those announcements, the Vera CPU announcement includes scientific computing among its target areas, but it presents a different configuration: a rack-scale platform combining CPUs and HPC networking. The materials do not describe how the products in these announcements work together or how their performance compares, so those relationships cannot be inferred.
The implication for the industry is that evaluations of scientific computing platforms will focus not only on GPU performance but also on the specifications of the entire system, including CPUs and networking. Since NVIDIA cited efficiency and throughput, future evaluations will need conditions and benchmarks to measure those gains. The announcement provides no such figures, however, and the company’s goals should be distinguished from measured results. Agentic science is also listed as a target use, raising questions about how models and agents will run on the platform, but no specific software configuration has been disclosed.
The next details to look for are the CPU’s core count and memory specifications, rack configuration, network bandwidth and connectivity method, power consumption and availability date. Without evaluation metrics for efficiency and throughput, the scientific computing workloads used and details on the platform’s configuration with GPUs, it will not be possible to compare its practical differences with existing HPC systems. At this point, there is no basis for discussing performance or deployment beyond the announced configuration and intended uses.
Details
According to NVIDIA, the platform is built at rack scale and combines the company’s first fully custom-designed data center CPU with networking technology that the company calls the “gold standard in HPC networking.” NVIDIA said it aims to improve efficiency and throughput for cutting-edge scientific modeling, simulation and agentic science.
The announcement materials do not specify the CPU, rack configuration, computing performance, power consumption or availability date.
Sources
Translated from the Japanese original: /news/nvidia-kagakukeisan-muke-vera-cpu-rakkusukeeru-kiban-wo-happ-e441c9. The Japanese version is authoritative if they differ.