Technical Intelligence Analyst Report

Date: 2026-03-04

Executive Summary

  • AMD EPYC Leadership in AI-RAN: Initial benchmarks of the newly established OCUDU project (5G/6G RAN) show AMD EPYC “Turin” processors outperforming Intel’s flagship “Granite Rapids” Xeon 6 servers in per-thread performance.
  • New Open Source Foundation: The Linux Foundation announced the OCUDU Ecosystem Foundation to advance open-source AI-RAN innovations, with premier members including AMD, NVIDIA, and SoftBank.
  • Telco-Optimized Silicon: The benchmark results suggest strong potential for AMD’s recently announced EPYC 8005 “Sorano” processors, which are specifically targeted at the Telco/RAN market.

🔲 AMD Hardware & Products

[2026-03-04] AMD EPYC Achieves Performance Leadership In New OCUDU Project For 5G/6G RAN

Source: Phoronix

Key takeaway relevant to AMD:

  • AMD EPYC 9005 “Turin” processors demonstrate superior per-thread performance compared to Intel’s flagship Xeon 6 “Granite Rapids” in Open RAN workloads.
  • The results validate AMD’s architecture for the 5G/6G telecommunications sector, particularly relevant for the upcoming EPYC 8005 “Sorano” series.
  • AMD is a premier member of the new OCUDU Ecosystem Foundation, securing a strategic position in the development of open-source AI-RAN stacks.

Summary:

  • The Linux Foundation announced the OCUDU Ecosystem Foundation at MWC, an evolution of the srsRAN project focused on 5G/6G CU (Centralized Unit) and DU (Distributed Unit) solutions.
  • Premier members include AMD, NVIDIA, Nokia, Ericsson, and SoftBank.
  • Phoronix conducted initial performance testing comparing current flagship AMD EPYC “Turin” servers against Intel Xeon 6 “Granite Rapids.”
  • Benchmarks reveal AMD maintains a performance lead in both PDSCH and PUSCH radio access network workloads.

Details:

  • Test Environment:
    • OS: Ubuntu 25.10 with Linux Kernel 6.18.
    • Storage: PCIe Gen 5 NVMe SSDs (identical across platforms).
    • Intel Hardware: Dual 128-core Intel Xeon 6980P (Granite Rapids) with maxed-out MRDIMM-8800 memory.
    • AMD Hardware: Dual 128-core EPYC 9755 “Turin” (Zen 5) and Dual 192-core EPYC 9965 “Turin” (Zen 5C).
  • Benchmark Specifics: Testing focused on per-thread performance due to a current 256-thread limit bug in the OCUDU total throughput test.
  • PDSCH (Physical Data Shared Channel) Results:
    • AMD EPYC 9755 (Zen 5): 11% faster per-thread than Intel Xeon 6980P.
    • AMD EPYC 9965 (Zen 5C): Matched the per-thread performance of the Xeon 6980P, despite being a density-optimized part.
  • PUSCH (Physical Uplink Shared Channel) Results:
    • AMD EPYC 9755 (Zen 5): ~20% faster per-thread than Intel Xeon 6980P.
    • AMD EPYC 9965 (Zen 5C): 4% faster per-thread than Intel Xeon 6980P.
  • Implications:
    • Even with Intel utilizing faster MRDIMM-8800 memory, AMD’s Zen 5 architecture holds a significant IPC advantage in these Telco workloads.
    • The “Dense” Zen 5C cores beating Intel’s Performance cores in PUSCH uplinks indicates high efficiency for high-density RAN deployments.
    • These results bode well for the EPYC 8005 “Sorano”, AMD’s specialized Telco/RAN CPU announced last week, and the future Zen 6 “Venice” architecture.
  • Timeline: OCUDU is working toward its first official release in April 2026.