Benchmark results published by Phoronix show that the NVIDIA Vera CPU delivers high throughput and memory bandwidth for agentic artificial intelligence workloads. The testing evaluated the processor’s capabilities in modern data center environments. Notably, the platform is designed to sustain performance when all cores are active.
The single-socket processor, rated at a 450-watt thermal design power, completed a default Linux kernel compilation in 20 seconds. This result represents a two-fold speed increase on a per-core basis compared to a 128-core x86 processor. Furthermore, the processor achieved a 10% higher performance score on a geometric mean basis than the AMD EPYC 9575F 5.0 GHz high-frequency processor.
Benchmark Results and Performance Metrics
In practical developer workloads, the platform demonstrated significant performance gains. The testing showed generational improvements across tasks including code compilation, file compression, video transcoding, and database management. These tasks represent the primary workloads executed by modern AI factories and software agents daily.
“The difference from Grace to Vera was consistently exceeding my expectations for gen-on-gen performance we typically see for processors.”
Michael Larabel, Founder of Phoronix
Architecture of the NVIDIA Vera CPU
The NVIDIA Vera CPU features 88 custom Olympus cores built on the Armv9.2 instruction set architecture. These cores handle sequential tasks such as branch-heavy runtimes, sandboxed code, data processing, and orchestration. Additionally, the monolithic die uses wide cores and the second-generation Scalable Coherency Fabric to manage data transfer across all active cores.
Memory Bandwidth and Efficiency
The NVIDIA Vera CPU integrates a second-generation LPDDR5X memory subsystem that provides up to 1.2 TB/s of bandwidth. This subsystem operates with less than 30 watts of memory power, compared to over 100 watts required by traditional DDR5 memory. In STREAM TRIAD testing, the platform sustained 90% of its peak memory bandwidth, which represents the highest percentage of rated peak bandwidth recorded in Phoronix database testing. Consequently, the architecture delivers over four times the memory bandwidth per core compared to traditional x86 processors.
Moreover, testing conducted by Prime Intellect showed that the processor maintains low and consistent memory latency. This stability remains constant even as multiple parallel workloads increase, providing predictable performance for complex software systems.
Market Availability and Deployment Options
Compared to the previous-generation Grace processor, the platform demonstrated a 1.6x geometric mean performance increase. The company has delivered the first units to selected cloud providers and artificial intelligence companies. Partners will offer the processors in single- and dual-socket configurations during the second half of 2026. These systems will support both air-cooled and liquid-cooled configurations for enterprise data centers.
