NVIDIA is framing Vera as a CPU designed for the changing demands of agentic Artificial Intelligence in the data center, where workloads depend on fast cores, high memory bandwidth and sustained performance across all active cores. The platform combines 88 NVIDIA custom Olympus cores, 1.2TB/s of memory bandwidth and an on-chip fabric intended to keep data moving efficiently while optimizing platform power. Olympus is fully compatible with the Armv9.2 instruction set architecture and is aimed at sequential CPU work such as branch-heavy runtimes, sandboxed code, data processing and orchestration.
Phoronix tested a single-socket Vera CPU rated at 450-watt thermal design power with less than 30 watts of memory power and reported strong results across code compilation, file compression, video transcoding, Python, Java and database management. NVIDIA links those workloads directly to the needs of agents and Artificial Intelligence factories, including compiling code, executing runtimes, compressing data, querying databases and coordinating large software stacks. Phoronix founder Michael Larabel described Vera as a significant competitive entry against Intel and AMD x86_64 processors.
Memory performance is a central part of NVIDIA’s argument for Vera. The chip uses a second-generation LPDDR5X memory subsystem that NVIDIA says enables lower energy per bit than DDR5. Vera can provide up to a massive 1.2 TB/s of bandwidth, up to 2x the peak memory bandwidth compared with traditional CPUs in less than 30 watts of memory power, as opposed to more than 100 watts for traditional DDR5. In Phoronix STREAM TRIAD testing, Vera sustained 90% of its peak memory bandwidth and delivered over 4x the memory bandwidth per core compared with traditional x86 CPUs. Separate testing by Prime Intellect found that Vera maintained high bandwidth and low, consistent memory latency as more workloads ran in parallel.
Phoronix testing also showed a large generational gain over Grace. Compared with the prior-generation NVIDIA Grace CPU, Vera delivered a 1.6x geometric mean increase in Phoronix’s testing. Vera led the tested CPU field, delivering a 1.5x overall performance advantage compared with a latest-generation 128-core x86 processor. Single-socket Vera compiled a default Linux kernel in just 20 seconds, and Vera delivered 2x faster Linux kernel compilation on a per-core basis compared with a 128-core processor. Larabel also wrote, “On a [geometric] mean basis, the NVIDIA Vera delivered 10% better performance than the AMD EPYC 9575F 5.0 GHz high frequency processor.”
NVIDIA said ecosystem support for Vera spans Artificial Intelligence natives, supercomputing centers, cloud service providers and infrastructure providers. The company has delivered the first Vera CPUs to leading Artificial Intelligence companies and cloud providers, with partner availability planned for the second half of the year. Vera will be offered by partners in dual- and single-socket systems, with air-cooled and liquid-cooled options for deployments ranging from enterprise data centers to high-density agentic Artificial Intelligence infrastructure.
