A benchmark review published October 6, 2026, reported that NVIDIA Vera led the overall composite across its tested HPC-focused workloads and took first place in a large-mesh OpenFOAM run against specified AMD EPYC Turin and Intel Xeon Granite Rapids systems. The tested two-socket Vera server paired 176 cores with LPDDR5-9600 memory.
What the Vera benchmark reports
NVIDIA Vera is an ARM64 server CPU, not a GPU. Each processor in the tested system had 88 cores; together, the two CPUs provided 176 cores and 352 threads. The reported results were particularly strong in several memory-intensive workloads.
The comparison covered more than one kind of computing task, including OpenFOAM, WRF, HPCG, LAMMPS, Xcompact3D, QuantLib, libxsmm and QMCPACK, as well as C-Ray, CoreMark and 7-Zip. That range matters: a strong result in one workload does not predict the ranking in every other one.
The systems and memory configurations compared
The test compared Vera with two AMD EPYC Turin configurations and a two-socket Intel Xeon Granite Rapids system. Their core counts and memory differed:
| System tested | CPU and count | Total cores/threads | Installed memory |
| NVIDIA Vera | 2 × NVIDIA Vera, 88 cores each | 176 cores / 352 threads | 16 × 96 GB Samsung LPDDR5-9600 |
| AMD EPYC 9575F | 2 × AMD EPYC 9575F | 128 cores / 256 threads | 24 × 64 GB Samsung DDR5-6400 |
| AMD EPYC 9755 | 2 × AMD EPYC 9755 | 256 cores / 512 threads | 24 × 64 GB Samsung DDR5-6400 |
| Intel Xeon 6980P | 2 × Intel Xeon 6980P | 256 cores / 512 threads | 24 × 64 GB Micron MRDIMM-8800 |
This is a comparison of complete systems, not an isolated measurement of CPU architecture: core counts and memory subsystems varied. Vera's configuration used LPDDR5-9600; the EPYC systems used DDR5-6400, while the Xeon system used MRDIMM-8800.
The software environments also differed. Vera ran a modified Ubuntu 24.04 LTS setup with a patched Linux 6.17-based kernel configured for 64K pages and GCC 15.3. The AMD and Intel systems used stock Ubuntu 26.04 LTS environments and a similar GCC 15-based compiler.
Why workload and mesh size matter
In the OpenFOAM drivaerFastback test, the leading system varied with mesh size and metric. Vera came in ahead of the Xeon 6980P and EPYC 9575F for small-mesh execution time, but behind the higher-core-count EPYC 9755. For the large mesh, Vera led the execution-time comparison and the performance-per-watt comparison.
The large-mesh result is a concrete win for Vera in that test condition—not a universal ranking across OpenFOAM workloads. Earlier coverage of Vera’s separate SMT test addressed a different comparison; this article covers HPC workloads against EPYC and Xeon systems.
CPU power in the reported test set
Across the HPC-focused tests, the two Vera CPUs averaged 842 W of CPU power and reached a measured peak of 1,044 W. Those readings describe the CPUs, not total server power. In the same test set, the reported typical peak frequency for Vera was 3.2–3.3 GHz, compared with 3.9 GHz for Xeon 6980P, 4.3 GHz for EPYC 9755 and 5.0 GHz for EPYC 9575F.
The review characterized the compared EPYC Turin and Xeon Granite Rapids generations as about two years old at publication. The measured outcomes therefore describe these systems and workloads, including Vera’s particularly strong results in selected memory-intensive tests.
