Nvidia's new Vera server processors outperformed top-tier AMD Epyc and Intel Xeon chips in high-performance computing benchmarks. The testing, conducted by Phoronix, utilized a dual-processor configuration featuring 176 cores and 352 threads.
The competitive analysis included dual-socket systems equipped with AMD Epyc 9575F and Epyc 9755 processors, as well as Intel Xeon 6980P chips from the Granite Rapids line. The Nvidia system demonstrated superior performance in memory-intensive workloads, specifically OpenFOAM CFD and WRF.
The performance advantage is attributed to the memory subsystem. Nvidia Vera utilizes LPDDR5x modules with a speed of 9600 MT/s, surpassing the DDR5-6400 memory used in AMD Epyc systems and the 8800 MT/s MRDIMM modules found in Intel Xeon 6 servers.
Although the Olympus Arm cores operate at a lower clock speed of 3.2–3.3 GHz compared to the 3.9–5.0 GHz range of x86 competitors, the processor delivered higher performance per clock. The system ran on Linux during the evaluation.
The power consumption for the dual Nvidia processor setup averaged 842 W, with peaks reaching 1044 W. This energy profile is comparable to the AMD Epyc 9755 and remains lower than the power requirements of the tested Intel systems.
