Phoronix has published the first public benchmark results for NVIDIA's Vera CPU, and the numbers show the chip performs strongly against established x86 processors from Intel and AMD. The benchmarks focused on the agentic AI workloads that Vera was built to handle in modern data centers.
Vera is powered by 88 custom NVIDIA Olympus cores, built on the Armv9.2 instruction set architecture. The chip delivers up to 1.2TB/s of memory bandwidth through a second-generation LPDDR5X memory subsystem, all within a 450-watt thermal design power envelope. The memory subsystem consumes less than 30 watts, compared to over 100 watts for traditional DDR5 memory.
Phoronix Benchmarks Reveal Strong CPU Performance
Michael Larabel, founder of Phoronix, ran a single-socket Vera CPU through a range of workloads common in AI factories: code compilation, file compression, video transcoding, Python, Java, and database management. He wrote that the results showed Vera is "the most formidable competition to Intel and AMD x86_64 processors ever realized."
Vera achieved a 1.5x overall performance advantage compared to a latest-generation 128-core x86 processor. In a geometric mean across all tests, Vera delivered 10% better performance than the AMD EPYC 9575F, which runs at 5.0 GHz. Single-socket Vera compiled a default Linux kernel in just 20 seconds, the fastest result Phoronix has ever measured in that test. On a per-core basis, Vera compiled the kernel 2x faster than the 128-core processor.
Memory Bandwidth and Efficiency Advantage
The LPDDR5X memory subsystem is a key differentiator. In Phoronix's STREAM TRIAD testing, Vera sustained 90% of its peak memory bandwidth, the highest percentage of any CPU Phoronix has tested. It delivered over 4x the memory bandwidth per core compared with traditional x86 CPUs.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Larabel noted: "NVIDIA Vera with its LPDDR5X memory was showing its incredible advantage in memory performance over current Intel Xeon and AMD EPYC processors."
Separate testing by Prime Intellect showed that Vera maintains high bandwidth and low, consistent memory latency as more workloads run in parallel, which is critical for agentic AI.
Generational Leap Over Grace
Compared to the previous-generation NVIDIA Grace CPU, Vera delivered a 1.6x geometric mean increase in Phoronix's testing. Larabel said the difference consistently exceeded his expectations for generation-over-generation performance gains typically seen in processors.
Customer Testing and Availability
NVIDIA announced at GTC that Vera has widespread ecosystem support from AI companies, supercomputing centers, cloud providers, and infrastructure providers. The first Vera CPUs have been delivered to leading AI companies and cloud providers, with partner availability expected in the second half of 2026. Partners will offer dual- and single-socket systems with air-cooled and liquid-cooled options for AI factory deployments.