Intel plans to launch its next generation’s Xeon platform, the Codename Oak Stream, in 2026, which will include Diamond Rapids, a CPU built for servers and workload with high performance.
Diamond Rapids use Intel’s 18A process and Panther Cove -Kerner, the same architecture that comes to future consumer chips.
The top model includes four calculation tiles, each with 48 performance kernels, adding a total of 192 cores per day. Socket.
New sock type
With support for single, double and quad-socket configurations, this means that a single rack could host up to 768 cores. That kind of density also brings challenges, especially with power. A fully configured rack could draw a full 2000W.
Intel says Diamond Rapids will use a new sock type, LGA 9324. It also supports modern standards such as PCIE 6.0 and CXL 3, which helps the processor connect to accelerators and fast storage.
Each CPU can support up to 16 channels of DDR5 memory and MRDIMM modules running at 12800 mt/s.
There will also be a version with fewer calculation tiles and a single I/O module that offers half the memory width and core count. This option may fit more power or space-conscious implementations.
Intel also throws this as a CPU ready for inference workload. While most AI treatments are still happening on GPUs, Diamond Rapids are aiming to improve CPU-based inference, especially for smaller models.
Native support for formats such as FP8 and TF32 should help in this room.
The CPU also supports Intel Apx and an upgraded AMX engine, two instruction sets designed to improve the effectiveness of modern server tasks. These upgrades suggest a future where CPUs can take on more AI workloads directly.
Diamond Rapids are expected to launch along with Intels Jaguar Shores AI Accelerator, part of his push to build a full AI-clear server platform.
Intel has not announced pricing yet and how it stacks up against competitors like AMD still seen.
Via Exprreview
