AMD's Venice server chips outpace Nvidia Vera in new SPEC benchmarks, claims up to 2.24x throughput
AMD published a white paper detailing SPEC CPU 2026 Integer Rate results for its EPYC Venice lineup, comparing against Nvidia's Vera and Intel's Xeon 6980P. The 256-core flagship reportedly delivers 2.24x the throughput of Vera and 2.37x that of Intel's top Xeon, while a 96-core high-frequency model is about 20% faster per-core than Vera. AMD notes configuration differences, including compiler versions, that may affect direct comparisons.
AMD's white paper compares its EPYC Venice lineup against Nvidia's Vera and Intel's Xeon 6980P using SPEC CPU 2026 Integer Rate tests. The 256-core flagship EPYC 9996 reportedly achieves 2.24x the throughput of Vera and 2.37x that of Intel's top Xeon, while also showing a 78% improvement over the previous-generation 192-core EPYC 9965.
The comparison includes notable caveats. AMD tested a down-cored 9996 (reduced from 256 to 96 cores) with a 600W power budget, and used GCC 16.1 compiler while Nvidia's published Vera results used GCC 15.2. Compiler version differences can significantly affect benchmark outcomes, with GCC 16 offering better optimizations but longer compile times.
These benchmark claims could reshape enterprise purchasing decisions in AI and cloud infrastructure, where server CPU throughput directly affects operational costs. If AMD's numbers hold under independent verification, organizations running large-scale compute workloads may see meaningful performance-per-dollar advantages, potentially accelerating adoption of AMD platforms. However, the compiler discrepancies and configuration differences mean buyers should await third-party validation before committing to large deployments, as real-world performance may vary from vendor-published figures.