Intel Xeon 6768P — 64-Core Granite Rapids-SP Server Processor

Specifications

Overview

The Intel Xeon 6768P is a 64-core Granite Rapids-SP processor on Intel's Socket 4710 platform, with 128 threads, 2.4 GHz base clock, 3.9 GHz max turbo and 336 MB of unified Smart Cache at 330 W. It is built on the Intel 3 process and is part of the Xeon 6 P-core family that replaced the Emerald Rapids generation in the mainstream 2-socket segment.

Memory is the headline change on this platform: eight DDR5 channels at 6400 MT/s give 409.6 GB/s, and the same sockets accept MRDIMMs running to 8000 MT/s (8800 MT/s on some SKUs) for a further bandwidth step. A 4 TB maximum memory capacity per socket is what lets a two-socket node carry the large in-memory datasets that AI inference on CPU, in-memory databases and electronic design automation all need. This part is the eight-socket-capable 64-core variant for scale-up platforms.

For AI-adjacent workloads the embedded Intel AMX accelerator matters more than clock speed: AMX provides matrix multiply in the core, which makes these chips viable for CPU-side inference and for pre/post-processing that would otherwise occupy a GPU. 88 PCIe 5.0 lanes and 64 CXL 2.0 lanes mean a two-socket board can hang eight accelerators and a full NVMe tier without a switch. QS Compute supplies the Xeon 6 range in volume with tray and boxed options.

Key Benefits

64 P-cores at 330 W — density for virtualisation and containerised workloads without dropping to E-core SKUs. MRDIMM-ready memory subsystem — 8-channel DDR5-6400 with MRDIMM support pushing toward 8000+ MT/s. AMX built in — usable CPU-side AI inference and preprocessing without a discrete accelerator.

Applications

Two-socket general-purpose and virtualisation servers, in-memory and relational databases, EDA and semiconductor design farms, CPU-side AI inference and data preprocessing, and storage controllers requiring high core counts with heavy PCIe fan-out.

Related Products