NVIDIA's Vera CPU, featuring custom Olympus cores and 1.2TB/s LPDDR5x memory bandwidth, is being deployed at HPC facilities such as Los Alamos. Recent testing evaluated the processor's performance in high-performance computing workloads, comparing it against current AMD EPYC and Intel Xeon offerings.
The test setup utilized a dual NVIDIA Vera server providing 176 cores and 352 threads, equipped with 16 modules of 96GB LPDDR5-9600 memory. The comparison included AMD EPYC 9575F and 9755 processors, as well as dual Intel Xeon 6980P chips, each configured with 24 memory modules.
For the software environment, the NVIDIA Vera server operated on a modified Ubuntu 24.04 LTS stack with a Linux 6.17-based kernel and GCC 15.3. The x86_64 systems ran on Ubuntu 26.04 LTS, utilizing a similar GCC 15-based compiler stack to ensure consistency across the benchmarks.
The evaluation process included monitoring real-time CPU power consumption and peak frequency metrics for each test. These benchmarks provide a technical comparison of the current generation of HPC processors from NVIDIA, AMD, and Intel.
