Intel has published the first MLPerf performance figures for its next-generation Xeon 6 CPUs, confirming that the Xeon 6900P chips will launch in September.
Intel declares huge AI performance gains with Intel 6 Xeon CPUs in MLPerf Inference v4.1; Granite Rapids-AP “Xeon 6900” coming to servers next month
In a recent press release, Intel showed off the AI performance of its upcoming Xeon 6 CPUs, known as Granite Rapids AP. Intel claims that the Xeon 6 processors can achieve up to 1.9x geometric mean performance improvement in AI inference compared to 5th Gen Xeon “Emerald Rapids” CPUs.
Additionally, the performance gains of Intel Xeon 6 processors are up to 17x higher than the 3rd Generation Xeon “Cascade Lake” series, which is an incredible feat for Intel to achieve over the past four years, but also somewhat expected given the new AI accelerators integrated into these chips and new instructions such as AVX-512 VNNI.
Intel has submitted the first six MLPerf benchmark results for its 5th Gen Xeons and the latest Granite Rapids “Xeon 6900P.” The MPerf Inference v4.1 tests are now publicly available on MLCommons and include results for the flagship Xeon 6 processor, the Xeon 6980P.
In MLPerf Inference v4.1, an industry-standard suite of AI performance benchmarks, Intel tested its processors on ResNet50, RetinaNet, 3DUNet, BERT, DLRM v2, and GPT-J. The Xeon 6980P delivered 1.9x higher AI inference performance across all of these benchmarks. Compared to third-generation processors released in 2021, it delivered approximately 17x performance on BERT and up to 15x performance on ResNet50.
“Intel is committed to delivering on this vision and is committed to delivering the best possible service to our customers,” said Pallavi Mahajan, corporate vice president and general manager of Data Center and AI Software, Intel.
“The latest MLPerf results demonstrate how critical continued investment and resource dedication is to advancing AI performance. Over the past four years, we have pushed the bar for AI performance on Xeon processors up to 17x, based on MLPerf. We look forward to working with our customers and partners to drive adoption of Xeon 6 as it nears general availability later this year.”
The Intel Xeon 6 6980P offers an impressive 128-core, 256-thread configuration with clock speeds ranging from 2.0GHz to 3.2GHz. The processor offers 504MB of L3 cache and has a TDP of 500W. Besides the 6980P, the series also has four other Granite Rapids AP chips with core/threads ranging from 72-core/144-thread to 120-core/240-thread. Intel has already launched the first Xeon 6 lineup, the Sierra Forest “Xeon 6700E” series, with the high-end P-Core family, the Xeon 6900P (128-core), due to launch in September, followed by the Xeon 6900E (288-core) and Xeon 6700P (86-core) in Q1 2025.
Such strong AI performance is crucial for Intel because it has five OEM partners — Cisco, Dell Technologies, HPE, Quanta and Supermicro — that build and sell computers and servers using Intel’s Xeon processors. Intel plans to release more information about the Granite Rapids AP “Xeon 6” CPUs at its official launch event in September.
