NVIDIA Vera Surpasses Intel and AMD Flagships in Server Benchmarks

The Phoronix lab has published benchmark results for Nvidia’s new Vera server processor based on the Arm architecture. The chip outperformed leading x86 server architectures from Intel and AMD, which have traditionally dominated the data center market.

Architecture and Its Features

Vera is Nvidia’s second server CPU after the Grace chip, but the first created specifically for AI agent workloads. The processor is optimized for orchestration, AI tool execution, data analysis, and related tasks.

The chip features:

• 88 specialized Arm v9.2 Olympus cores with 176 threads.
• Native FP8 support.
• Memory bandwidth of 1.2 TB/s with support for up to 1.5 TB of LPDDR5X memory in the SOCAMM2 form factor.
• A second-generation Scalable Coherency Fabric interconnect delivering 3.4 TB/s bandwidth to reduce latency.

Dominating the Benchmarks

During testing, Nvidia Vera was compared against the Intel Xeon Granite Rapids 6980P and the AMD EPYC Turin lineup. The benchmark suite included code compilation, video encoding, memory streaming performance, database workloads, and Python/Java processing.

Nvidia Vera led across all benchmarks. It outperformed AMD Turin by nearly 11% and exceeded single-socket Intel Xeon configurations by 55.3%.

Energy Efficiency and Future Revenue Potential

With a thermal design power of 450W for the processor itself and 50W for a 768 GB memory pool, Vera delivers industry-leading energy efficiency among Arm architectures in its class.

This technological leap could become a major success for Nvidia. According to analyst forecasts, sales of Vera and Grace processors could generate approximately $20 billion in revenue for the company.