Since receiving the Gigabyte R284-A92-AAL1 a while back as a Xeon 6900 series 2U server platform to replace the failed Intel AvenueCity reference server, I have been getting caught-up in fresh Xeon 6980P Granite Rapids benchmarks with the latest software updates over the past year. I’ve provided fresh looks at the DDR5-6400 vs. MRDIMM-8800 performance, the AMX benefits for AI, SNC3 vs. HEX mode, Latency Optimized Mode, Cache Aware Scheduling, and more with the fresh Linux software stack and this production Gigabyte server platform. One of the areas I have been meaning to re-visit is a fresh head-to-head benchmark battle between 5th Gen AMD EPYC “Turin” and Intel Xeon 6 “Granite Rapids”. In this article is a 128-core showdown between the Xeon 6980P and EPYC 9755 128-core processors with the latest open-source Linux software as of the end of 2025.
As 2025 comes to an end, today’s article is a fresh look at these latest-generation Intel Xeon and AMD EPYC processors while using an up-to-date Linux software stack compared to the prior launch day testing last year. The AMD EPYC 9755 and Xeon 6980P both are 128 core / 256 thread parts, or 256 cores / 512 threads in total with the dual socket testing being done today. These processors were used for comparing at the same core/thread count. The Intel Xeon 6 and AMD EPYC 9005 series can go higher with the Sierra Forest and Zen 5(C) Dense CPU options but for this article are looking squarely at a battle with the full “performance” cores. There will be some fresh EPYC 9965 benchmarks soon on Phoronix for those interested in seeing fresh numbers there for that 192 core / 384 thread processor.
Ubuntu 25.10 was run on both the Intel and AMD servers while upgrading to the Linux 6.18 kernel for an up-to-date kernel with the latest upstream innovations there. GCC 15.2 was in use as the latest stable GNU compiler. Both servers were running with the performance governor during the testing.
The Intel Xeon 6980P 2P server was using the Gigabyte R284-A92-AAL1 with 24 x 64GB MRDIMM-8800 memory modules. The AMD EPYC 9755 2P testing was on the AMD Volcano reference server with the 24 x 64GB DDR5-6400 Samsung M321R8GA0PB1-CCPKC memory modules. With a Volcano BIOS update since launch, the memory speed was lifted from DDR5-6000 to DDR5-6400 support for this platform. The AMD Volcano reference server was still used for lack of any other AMD EPYC 9005 series 2P 500W production server platform in the lab.
Besides looking at the raw CPU performance, the CPU power consumption was also monitored on a per-test basis too for being able to analyze the power efficiency between these 128 core Turin and Granite Rapids processors. Due to the Intel server being a production platform from an ODM and the AMD server being a pre-production reference server, the overall AC system power consumption of the entire servers were not compared: pre-production servers typically aren’t tuned for power efficiency / fine-tuned fan curves and the like with both AMD and Intel having recommended in the past not to bother comparing pre-production server power consumption numbers. Thus due to that constraint just comparing the combined 2P CPU power consumption.
Nearly 200 benchmarks were conducted spanning a variety of different domains/workloads in seeing how these 128 core Intel Granite Rapids and AMD EPYC Turin processors are comparing now one year post-launch and running on the latest upstream Linux 6.18 kernel and Ubuntu 25.10, which is becoming quite close to the state of Ubuntu 26.04 LTS next April.
