Intel Launches Xeon 6900P, 128 P-Core Granite Rapids Arrives To Battle AMD EPYC

The Xeon 6 6900P series doesn’t only offer more cores per socket and higher performance than the previous-gen, but it’s also more efficient as well. In fact, Intel is claiming higher efficiency with the Xeon 6 6900P series versus the previous-gen across the range of server utilization.


Of course, the chart supplied compares only a single configuration, and this isn’t data we’ve generated on our own, but Intel is claiming up to 1.9x performance per watt advantage for the 128 core / 500W TDP Xeon 6 6980P versus the 64 core, 350W 5th Gen Xeon 8592+. Since most servers aren’t running at full tilt, 24/7, and typical utilization targets hover in the 50% mark -- give or take -- Intel has tuned the 6900P’s power management scheme for max efficiency in the middle range. In fact, the power optimized mode that was an option on previous-gen Xeons is now on by default with the Xeon 6 6900P.

Xeon 6 6900P Expected Performance

At similar core counts, Intel is also claiming some massive performance gains versus the previous-gen Emerald Rapids Xeons and AMD’s EPYC processors, especially with AI workloads. Across and array of models and use cases, Intel’s data shows the 96 core Xeon 6 6972P dominating the AMD EPYC 9654 in AI inference performance. The 5th Gen Xeon 8592+ tops out at "only" 64 cores, so it obviously can't match the Xeon 6 here either.


Across an array of general compute, web service, HPC and AI workloads, Intel is also claiming large raw performance and performance per watt advantages versus the previous-gen. The chart above is a conglomeration of results from different 6900P configurations versus the 5th Gen Xeon 8592+ and there’s simply no comparison. The Xeon 6 6900P’s increased core counts in improved efficiency provide big wins across the board.

Intel Xeon 6 6900P All P-Core Processors: Initial Conclusion

As we’ve mentioned in the past, Intel is trying to address as many potential customers as possible with its Xeon 6 series of processors. Whether a customer needs massive core counts for web and micro-service workloads or the highest performance cores for AI and deep learning, Intel’s got a potential solution.


There are still a number of additional products in the Xeon 6 pipeline, due to arrive later this year and into 2025, in both P and E-core varieties. In the end though, it seems Intel is headed in the right direction with the Xeon 6 family. Power efficiency and perf-per-watt has been improved, per-socket core density has increased considerably (with both E-core and P-core based models), and performance should be significantly better in the vast majority of workloads, especially AI and HPC.

We hope to get our hands on a Intel Xeon 6 6900P-based server soon, to provide some fresh data of our own. For now though, Intel’s benchmark data shows some big performance and efficiency gains across an array of workloads and use cases, which is just what the data center market is clamoring for currently.