Initial benchmarks of NVIDIA's Vera CPU, published by Phoronix and reported by Tom's Hardware and HPCwire, show the...

Initial benchmarks of NVIDIA's Vera CPU, published by Phoronix and reported by Tom's Hardware and HPCwire, show the 88-core ARM-based processor matching or beating AMD's EPYC 'Turin' chips in key server workloads, including code compilation, database performance, and AI agent orchestration Source 1 Source 5. The Vera CPU delivers up to 2x memory bandwidth vs. x86 rivals and sustained 90% of peak STREAM bandwidth, a metric where Phoronix notes it 'absolutely dusted' AMD EPYC. NVIDIA's first-generation custom server core, codenamed Olympus, achieved a 1.6x geometric mean performance uplift over its predecessor Grace, and in per-core metrics like Linux kernel compilation, Vera topped the stack. While the benchmarks were curated for Vera's target AI agent workloads, the results represent 'the most formidable competition to Intel and AMD x86_64 processors ever realized,' per Phoronix. This development threatens AMD's recent server market share gains, particularly as hyperscalers evaluate Vera for AI inference and agentic AI tasks, where its memory bandwidth and core count align closely with workload requirements. AMD's EPYC architectures (Turin) which had been competitive on core count and bandwidth now face a credible ARM-based alternative from NVIDIA, which also bundles its own GPUs and networking. The risk is that key customers like cloud providers may consolidate on NVIDIA's full-stack ecosystem, eroding AMD's design wins. Cisco and other server OEMs have already indicated plans to support Vera. Given Vera's 450W TDP and availability—NVIDIA is providing early access to partners—this could pressure AMD's EPYC roadmap in the 12-18 month window as Vera ramps to production. Spillover effects include Intel's Xeon, which also trails Vera in these tests, and the broader x86 server ecosystem, while NVIDIA's competitive moat widens.