Intel plans to launch its next -generation XEON platform, Codenamed Oak Stream, in 2026, which will include Diamond Rapids, a CPU built for servers and high -performance work loads.
Diamond Rapids will use the 18th process of Intel and the Cove Cove nuclei, the same architecture reaching future consumption chips.
The upper model includes four computing mosaics, each with 48 performance cores, adding to a total of 192 cores per plug.
New type of plug
With support for single, dual and quadruple configurations, that means that a single shelf could house up to 768 cores. That type of density also brings challenges, particularly with power. A totally configured frame could draw 2000W.
Intel says that Diamond Rapid will use a new type of socket, LGA 9324. It also admits modern standards such as PCIE 6.0 and CXL 3, helping the processor to connect to accelerators and rapid storage.
Each CPU can admit up to 16 DDR5 memory channels and MRDIMM modules that are executed at 12800mt/s.
There will also be a version with less computing mosaics and a single I/O module, which offers half the memory bandwidth and the nucleus count. This option can adapt to more conscious implementations of energy or spaces.
Intel is also launching this as a CPU ready for inference work loads. Although most of the IA processing still occurs in the GPU, Diamond Rapids aims to improve CPU -based inference, especially for smaller models.
Native support for formats such as FP8 and TF32 should help in that space.
The CPU will also admit Intel APX and an updated AMX engine, two sets of instructions aimed at improving efficiency for modern server tasks. These updates hint at a future in which CPUs can assume more workloads directly.
Diamond Rapid is expected to be launched together with the Jaguar Shores AI AI accelerator, part of its impulse to build a complete list.
Intel has not yet announced prices, and it remains to be seen how it compares with competitors like AMD.
Through squeeze
