Nvidia $NVDA published detailed specifications and benchmark results for its Vera data center CPU on Tuesday, giving prospective customers the architectural detail needed to evaluate the chip against processors from AMD $AMD and Intel $INTC.
The chip maker published a white paper and SPEC CPU 2026 results showing Vera ahead of AMD's Epyc 9755 in integer performance
PATRICK T. FALLON / Getty Images
Nvidia $NVDA published detailed specifications and benchmark results for its Vera data center CPU on Tuesday, giving prospective customers the architectural detail needed to evaluate the chip against processors from AMD $AMD and Intel $INTC.
The company released a white paper alongside unofficial SPEC CPU 2026 results comparing Vera against AMD's Epyc 9755, with both chips running in a dual-socket configuration, according to Tom's Hardware. In the SPECrate integer suite, Vera's overall base score came to 925, edging out 898 for a dual-socket Epyc 9755 system — a roughly 3% margin that Vera achieved while operating with a smaller thread count. Nvidia compiled both test systems with GNU 15.2.
Vera is built around Nvidia's custom Olympus core, which the company designed to prioritize single-thread performance and memory bandwidth rather than raw core density. The chip packs 88 Olympus cores and 176 simultaneous multithreading threads, the company said. Its memory subsystem delivers up to 1.2 terabytes per second of aggregate bandwidth using SOCAMM2 LPDDR5X modules, and an on-die coherency fabric provides up to 3.4 TB/s of internal fabric bandwidth alongside a 164 MB unified L3 cache.
Nvidia said the chip was designed for agentic AI workloads, which depend on strong per-thread progress through branch-heavy code, pointer-heavy data structures, and long dependency chains. Cloud platform DeepInfra ran independent benchmarks showing Vera supporting up to 1.6 times more concurrent AI agents and up to 2.2 times faster orchestration than alternative CPUs, the company said.
Nvidia said Vera chips were delivered to clients including OpenAI, Anthropic, and SpaceX in June, according to CNBC. Sales options include a standalone chip, a liquid-cooled 256-chip rack, a dual-chip single-server configuration, and a bundled offering with Nvidia's Rubin GPUs under the Vera Rubin platform name. The chip has a thermal design range of 250 to 450 watts and can address as much as 1.5 terabytes of memory.
Nvidia enters the server CPU market as a challenger. According to CNBC, AMD commands around 33% of the server CPU market while Intel holds approximately 66.8%, and each has established strong relationships with major cloud and hyperscale customers. At a presentation the previous week, Nvidia hyperscale vice president Ian Buck argued that the rise of agentic AI had made CPUs "much more integral," singling out "how fast a CPU can answer one question." Nvidia estimated the total server CPU market could eventually reach $200 billion.
Nvidia confirmed at the Computex conference in June that the Vera CPU had entered full production, with the chip packing 88 cores and an LPDDR5X memory subsystem capable of 1.2 TB/s of bandwidth. A separate Computex announcement saw Hewlett Packard Enterprise unveil a Vera-powered server, the ProLiant Compute DL394 Gen12, targeted at agentic AI orchestration and reinforcement learning workloads, with availability expected in fall 2026.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.