NVIDIA's Arm-based 'Vera' CPU benchmarked, beats Intel Xeon and AMD EPYC CPUs

NVIDIA's Vera CPU has been making headlines ever since it was announced at GTC 2026, and Phoronix has finally put some real numbers behind the hype. The publication was given early access to pre-production Vera hardware and ran it through a permitted subset of workloads, including code compilation, STREAM memory benchmarks, video encoding, database performance, and Python and Java tasks.

Under the hood, Vera is built around 88 custom Armv9.2 "Olympus" cores on a unified monolithic die, delivering 176 threads through physical resource partitioning. The chip supports native FP8 processing for AI workloads, pairs with up to 1.5 TB of LPDDR5X memory in the SOCAMM2 form factor, and hits 1.2 TB/s of memory bandwidth. A second-generation Scalable Coherency Fabric provides 3.4 TB/s of bisection bandwidth, which helps sidestep the latency issues that typically come with chiplet-based designs.

3

The benchmark results are impressive. Across the geometric mean of all tested workloads, Vera came out nearly 11% ahead of AMD's top EPYC offerings and around 55% ahead of the best single-socket Intel Xeon. From generation to generation, Vera clocked in at roughly 1.63 times the performance of NVIDIA's previous Grace CPU. Phoronix concluded that this is the most capable ARM-based Linux server processor it has ever tested, which is a statement worth sitting with for a moment.

All of this lines up with what we covered earlier this week. NVIDIA has already hand-delivered its first Vera units to partners, including Anthropic, OpenAI, SpaceX, and Oracle. NVIDIA's CFO, Colette Kress, has projected that the company's CPU lineup could generate around $20 billion in revenue in the current fiscal year, with NVIDIA openly targeting the position of the world's leading CPU supplier.

3

We also reported that analysts at GF Securities expect NVIDIA to formally showcase Vera at Computex 2026, where the company is expected to claim the chip delivers around 1.5 times the inference performance of competing x86 offerings, alongside double the overall throughput. With the new benchmark data, it looks like Vera is on track to meet those lofty goals.

It is worth noting that per-watt performance data was not part of this first round of testing, as NVIDIA did not permit those benchmarks on the pre-production hardware. That is one metric to keep an eye on when full production units become available.

Looking ahead, the competition is not standing still. AMD's next-gen EPYC Venice, based on the Zen 6 architecture, is already in mass production and targeting a second-half 2026 release, while Intel is preparing its Diamond Rapids platform in response. For now, though, the early CPU performance crown belongs to NVIDIA, which is a sentence that I was not expecting to write 4 years ago.