Welcome Here  Shenzhen Mingjiada Electronics Co., Ltd.

On 7 July, Intel announced that it had launched the Xeon 6 performance-core and efficiency-core processors Granite Rapids and Sierra Forest, both based on the Intel 3 process technology. The platform codename for Intel's next-generation Xeon processors is Oak Stream, which includes Diamond Rapids using performance cores and Clearwater Forest using efficiency and efficiency cores.

According to HEPiX TechWatch, the top-of-the-line model of Diamond Rapids will feature up to 192 P-cores, distributed across four compute modules, with 48 cores per module. Each pair of compute modules is connected to an I/O module, Two I/O modules are interconnected via a bus, providing up to 16 DDR5 memory channels. Additionally, it supports PCIe 6.0, CXL 3, and UPI 3, with a processor TDP of 500W.

There is also a version with a single I/O module, supporting only 8 memory channels, essentially halving the top-tier processor. The next-generation Xeon processors will support MRDIMM memory, with a maximum memory speed of 12,800 MT/s, and 16 memory channels providing extremely high bandwidth.

Diamond Rapids will use the Panther Cove core, which is the P-core of the upcoming Panther Lake processor, and is planned to be manufactured using Intel's 18A process. Diamond Rapids will be the first to support Intel APX and significantly improve AMX efficiency. They also plan to embed native support for more floating-point formats, such as TF32 and FP8. In fact, most inference models can now run on CPUs, and Intel aims to accelerate the basic inference speed of small models, allowing users to run inference workloads directly on CPUs.

Diamond Rapids uses the LGA 9324 interface, supporting single-socket, dual-socket, and quad-socket topologies, capable of providing up to 768 cores in a single server rack, with a power consumption of 2000W. It also supports PCIe 6.0 for expanding other accelerators. Intel plans to launch this product in 2026 and may simultaneously introduce the Jaguar Shores AI accelerator to create a complete AI system.