Intel Xeon 6980P "Granite Rapids" Linux Benchmarks
When taking the geometric mean of all the benchmarks that ran successfully on all of the CPUs (while removing the Sierra Forest CPUs since they failed to run a few of the HPC benchmarks), here is where the Xeon 6980P dual socket configuration stands for launch day. The Xeon 6980P does manage to land well in front of the AMD EPYC 9684X flagship Genoa-X SKU overall along with the EPYC 9754 Bergamo processors. Generationally around 1.38x the performance with Granite Rapids compared to the prior Emerald Rapids flagships granted there is twice the number of cores.
Again, follow-up articles will be looking at the CPU power consumption / performance-per-Watt due to a kernel issue preventing accurate RAPL/PowerCap readings on Granite Rapids during the initial run and the limited amount of time for testing ahead of launch day.
Granite Rapids has exceeded my expectations going into this testing and especially for AI and HPC / technical computing workloads is a very strong performance. The uplift in some of the high performance computing benchmarks was huge over Emerald Rapids and the AMD EPYC competition. A number of those HPC wins are likely driven in large part by the introduction of MRDIMM memory at 8800 MT/s. Again, a follow-up article will be quantifying the Granite Rapids performance difference between DDR5-6400 and MRDIMMs.
With the Xeon 6900P series launch Intel has shown they can return to competing with AMD EPYC at the top-end of HPC performance and other compute intensive workloads. In areas of AI where the software is optimized for Advanced Matrix Extensions (AMX), the leads with Granite Rapids are extended further compared to prior generations.
CXL 2.0 support, increased L3 cache sizes, and other improvements round out the Xeon 6 Granite Rapids offerings. There also remains the integrated IP accelerators with Granite Rapids but no major improvements there and as noted the security issues for assigning accelerators securely to VMs haven't been addressed for this generation. Granite Rapids also still tops out at only up to 96 lanes of PCIe 5.0 compared to 128 lanes with the latest AMD EPYC processors.
👁 Intel AvenueCity OpenBMC with Xeon 6980P CPUs
Coming up in a few weeks AMD is upping their game with the 5th Gen AMD EPYC "Turin" launch so it will be very interesting to see how those next-gen server processors are geared to compete for performance and power efficiency. It will also be interesting to see how they compete in value/TCO, especially with so far not seeing any Xeon 6900P series pricing. Similarly, how robust the actual availability is in the near-term for these new processors and server platforms for Turin and Granite Rapids.
As far as the Linux support goes for Intel Xeon 6 Granite Rapids, it is in great shape as we've come to expect and love over the years. Intel does a stellar job ensuring all of the necessary support is upstream in the Linux kernel and related components well ahead of launch. I continue to be impressed by Intel's timely upstream enabling of new hardware support and equally as important is all the work they continue investing into many open-source projects for ensuring optimal performance. The most stark difference in the AMD vs. Intel software support approach continues to be around the compilers... Intel began working on AMX-FP16 for the compilers in late 2022, Granite Rapids GCC support came in November 2022, and followed by the LLVM Clang target. Meanwhile AMD only added Zen 5 (znver5) support to GCC earlier in the year and literally just days ago to the LLVM/Clang compiler. For these competing server processors launching at roughly the same time, Intel had their GCC and LLVM/Clang support upstreamed nearly two years prior. This is important with GCC issuing new feature releases annual and LLVM on a half-year regiment while Linux distributions also are typically conservative in moving to new compiler releases (e.g. only Ubuntu 24.10 seeing GCC 14). In turn this allows Intel Granite Rapids HPC customers and others seeking optimized compiler support with "-march=graniterapids" to comfortable use the LLVM/Clang and GCC compilers found out-of-the-box in Ubuntu 24.04 LTS and other modern Linux distributions. AMD znver5 targeting won't enjoy that same level of out-of-the-box compiler support in Linux distributions now until beginning to hit more 2025 Linux distributions.
The Intel Xeon 6900P Granite Rapids performance has been very exciting to explore and will continue to be looked at in different areas in follow-up articles over the coming weeks on Phoronix. Thanks to Intel for providing the Granite Rapids processors with Avenue City reference server for the launch-day testing.
If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.