NVIDIA's Vera ARM CPU beats Intel Xeon and AMD EPYC in early benchmarks
The most disruptive move in server CPUs isn't a faster x86 chip — it's an ARM chip from a GPU company that just beat the incumbents in their own game.
Get AI news that actually matters
One email a day. Zero fluff. Join 10,000+ professionals.
NVIDIA's Vera CPU, built around 88 custom Olympus ARM cores, has reportedly outperformed Intel's upcoming Granite Rapids and AMD's EPYC Turin in early server benchmarks. While the results come from limited internal or early-access testing and await independent verification, they mark a pivotal moment: a GPU-centric company is now leading in general-purpose CPU performance for the first time. Vera is designed not just for standalone servers but as the compute backbone for NVIDIA's next-generation Rubin AI platform, effectively extending the company's dominance from GPUs to the entire server node.
The competitive landscape is now defined by three architectural camps. Intel and AMD have long dominated with x86, leveraging decades of software compatibility and massive installed bases. Intel's Granite Rapids focuses on performance cores for AI and HPC, while AMD's EPYC Turin brings high core counts and chiplets. On the ARM side, Ampere Computing has championed cloud-native efficiency but lacks the raw single-thread and AI-tailored performance Vera promises. NVIDIA's entry with custom cores optimized for AI workloads — rather than generic ARM designs — creates a unique hybrid: a CPU that acts as a coordinated partner to its own GPUs, communicating over NVLink and other proprietary interconnects. This integration is Vera's real weapon, not just raw SPEC scores.
The broader implication is a fundamental shift in semiconductor strategy. NVIDIA is no longer a GPU supplier; it is building a complete, vertically integrated computing platform — from CPU and GPU to networking (Mellanox) and data processing units (DPUs). This approach threatens to lock customers into an NVIDIA ecosystem where every component is optimized together, potentially reducing the role of x86 partners like Intel and AMD in AI infrastructure. The server CPU market, worth roughly $100 billion annually, becomes a new addressable market for NVIDIA, though initial volumes will likely target high-end AI clusters. However, the risks are substantial: ARM software compatibility for legacy enterprise applications remains a barrier, power and thermal characteristics of 88-core designs are unproven at scale, and early hype could backfire if third-party validation disappoints.
The bottom line: Vera's early benchmark lead is less about raw performance and more about architectural ambition. NVIDIA is betting that the future of compute is specialized, integrated, and tied to a single provider — a stark contrast to the modular, open-x86 ethos that built the modern data center. If Vera succeeds, it won't just displace Intel and AMD; it will redefine how servers are built and sold.
- NVIDIA's Vera CPU with 88 custom ARM cores outperforms top x86 chips in early benchmarks, signaling ARM's datacenter credibility beyond cloud instances.
- By integrating Vera with Rubin GPUs and proprietary interconnects, NVIDIA creates a locked-in AI platform that reduces dependence on Intel and AMD components.
- Success depends on third-party validation and ARM software ecosystem maturity; x86 incumbents still dominate legacy enterprise workloads.
Why It Matters
NVIDIA is redefining server architecture by combining its own CPU and GPU, challenging x86 dominance in a $100B market.