Intel plans to launch its next generation XEON platform, codenmed oak stream in 2026, including Diamond Rapids, a CPU designed for server and high-performance workloads.
Diamond Rapids will use the 18A process of Intel and Panther Cove Core, which is the same architecture in future consumer chips.
The top model consists of four computable tiles, each of which has 48 performance cores, adding up to 192 core per socket.
New socket type
With support for single, double-, and quad-clat configuration, it means that a single rack can host up to 768 core. Such density also brings challenges, especially with power. A fully configured rack can do 2000W one more.
Intel says that Diamond Rapids will use a new socket type, LGA 9324. It also supports modern standards such as PCI 6.0 and CXL3, which helps the processor connect to accelerator and fast storage.
Each CPU can support DDR5 memory and 16 channels of MRDIMM modules running at 12800mt/s.
There will also be a version with low compute tiles and a single I/O module, offering half a memory bandwidth and core count. That option may correspond to more power- or space-conscious deployment.
Intel is also pitching it as a CPU ready for workload. While most AI processing still occurs on the GPU, Diamond Rapids aims to improve CPU-based estimates, especially for small models.
The native support for formats like FP8 and TF32 should help in that location.
The CPU will also support Intel APX and an advanced AMX engine, two instructions set to improve efficiency for modern server functions. These upgrades indicate in a future where CPUs can directly take directly to AI workload.
Diamond Rapids is expected to launch Intel’s Jaguar Shorus AI Accelerator, which is part of its push to create a full A-Taiyar server platform.
Intel has not yet announced pricing, and how it piles up against competitors such as AMD.
Through Boycott
