# Phoronix benchmarks AMD EPYC 9006 Venice against NVIDIA Vera for HPC workloads

> **Key takeaway:** AMD's EPYC 9006 Venice server processors face NVIDIA's Vera CPU in head-to-head HPC benchmarks on Linux 7.3 with GCC 15, with power and frequency telemetry collected throughout.

**Published:** 2026-11-11T15:44:41+00:00 
**Source:** Phoronix Benchmarks & Drivers 
**Category:** kernel-os 
**Canonical URL:** https://fosswire.org/news/phoronix-benchmarks-amd-epyc-9006-venice-against-nvidia-vera.html 

## Summary
Michael Larabel at Phoronix has published a dedicated comparison pitting AMD's newly launched EPYC 9006 Venice server processors against NVIDIA's Vera CPU across multiple HPC workloads. The testing focused on two Venice parts: the 256-core EPYC 9996 and the 96-core high-frequency EPYC 9686F. Because Vera access was limited to prior benchmarking sessions, Larabel re-ran Venice tests on Ubuntu 26.10 with the Linux 7.3 kernel and GCC 15 to match the software stack used in the earlier Vera article, rather than using the Ubuntu 26.04 LTS configuration from Venice's launch-day coverage.

The benchmarks span several pages of HPC-oriented tests, with CPU frequency and peak power consumption monitored throughout. Notably, GCC 15 was selected as the default compiler for parity with the Vera setup, while GCC 16—which introduces Znver6 tuning for Zen 6—was reserved for future dedicated compiler benchmarking articles. This comparison arrives as both AMD and NVIDIA push into the high-core-count server market, with Venice on the new SP7 socket and Vera representing NVIDIA's first CPU entry into the x86 server space.

## Benchmark methodology and hardware context
- **Kernel and platform:** Venice tests were run on Ubuntu 26.10 with the Linux 7.3 kernel, a newer stack than the Ubuntu 26.04 LTS used in Venice launch-day reviews, to match Vera's prior benchmarking environment.
- **Compiler parity:** GCC 15 served as the default compiler for both platforms; GCC 16 with Znver6 support was explicitly excluded to maintain comparison validity and will be tested separately.
- **Hardware under test:** The comparison covers AMD's EPYC 9996 (256 cores) and EPYC 9686F (96 cores, high frequency) against NVIDIA Vera, with both raw performance and power/telemetry data collected.
- **Data reuse constraint:** Vera numbers were drawn from earlier benchmarking sessions due to lack of persistent remote or physical access, necessitating fresh Venice runs on the matched software stack.
- **Telemetry scope:** Beyond raw HPC workload throughput, peak CPU frequency and CPU power consumption were monitored continuously throughout the benchmark process.

## Why it matters
This head-to-head comparison provides rare empirical data on NVIDIA's Vera CPU against established x86 server silicon under Linux, informing procurement decisions for HPC clusters and cloud infrastructure. For the open-source ecosystem, the matched use of GCC 15 and Linux 7.3 demonstrates how compiler and kernel versioning choices directly affect competitive benchmarking validity, while the deferred GCC 16/Znver6 testing signals that compiler optimization maturity for Zen 6 remains an evolving story.