Intel Unveils Clearwater Forest Xeon CPU: Next-Gen 12 Chiplet Design, 288 Darkmont E-Cores, 17% IPC Boost, 2x L2 Cache Bandwidth, and DDR5-8000 Support

Intel has recently revealed groundbreaking details about its upcoming Clearwater Forest “E-Core”Xeon CPUs, which will boast a staggering 288 cores and be constructed using the advanced 18A process node. This marks an important development in the realm of server technology.

Introduction of Intel’s Next Generation Clearwater Forest E-Core Xeon CPUs

The Clearwater Forest series of Xeon CPUs represents Intel’s commitment to enhancing server performance with an E-Core configuration. Similar to the segmentation seen in the Xeon 6 family—comprising both P-Core and E-Core options like Granite Rapids and Sierra Forest—these new CPUs will be classified into two distinct lineups: the P-Core focused “Diamond Rapids”and the E-Core specific “Clearwater Forest.”The P-Core lineup is designed to excel in high-performance scenarios, particularly for demanding AI and compute-intensive tasks, while the E-Core lineup aims for greater efficiency, particularly suitable for high-density scale-out workloads.

Intel Xeon 6 processor performance and efficiency optimization overview.

Exciting Features of Clearwater Forest Xeon CPUs

During the Hot Chips 2025 conference, Intel disclosed vital insights regarding the Clearwater Forest Xeon CPUs, built on the cutting-edge 18A process node, which is also being utilized in the upcoming Panther Lake CPUs expected to launch later this year. Here are some key highlights of the new Xeon E-Core CPUs:

Advanced 18A Process Node: Promises enhanced performance and power efficiency.
New Efficiency Core Architecture: Inter-Processor Communication (IPC) improvements tailored for the 18A process.
Foveros Direct 3D Construction: Facilitates shorter, energy-efficient routes and a larger last-level cache (LLC).
Boosted Memory Bandwidth: Supports 12-Channel DDR5-8000 RAM.

Intel Clearwater Forest chip showcasing efficiency in data center applications.

Insights into the Architecture of Clearwater Forest

At the core of Clearwater Forest lies Intel’s Darkmont E-Core architecture, an evolution from the Sierra Glen E-Cores. This design offers numerous enhancements targeted at optimizing performance:

Refined Front-End Structure
Enhanced Out-of-Order Execution (OOE) Engine
Expansion of Scalar and Vector Execution Resources
Improved Memory Subsystem

Clearwater Forest E-core architecture with optimized CPU subsystems.

Performance Enhancements in Instruction and Execution

The enhanced front end boasts a 64kB instruction cache and three 3-wide instruction decoders, delivering a remarkable 50% increase in instruction bandwidth, supporting up to nine decodes per cycle, aided by a more precise branch predictor leveraging deep branch histories.

Out-of-Order Engine diagram of Clearwater Forest illustrating execution parallelism features.

The OOE Engine has been significantly upgraded, featuring an 8-wide allocation and a 2x increase in 16-wide retire capabilities. The capacity for out-of-order execution also sees a 60% enhancement with 416 units. Moreover, the 26 execution ports now offer a 50% increase in efficiency compared to the previous generation.

Memory System and Additional Features

The memory subsystem includes advancements giving it a 50% increase in three-load capabilities while maintaining two-load features. Additionally, load issuing has been made earlier to curtail latency, and there is enhanced support for up to 128 outstanding L2 misses—doubling previous capacities. Unique characteristics of the Clearwater Forest Xeon CPUs include:

ECC for L1 Data Cache
Data Poisoning Support
Recoverable Machine Check Capability
Local Machine Check Functions
52 Physical Address Bits
Core Lockstep Configuration

Diagram of Clearwater Forest E-core memory subsystem detailing advanced features.

Cutting-Edge Modular Architecture

The Clearwater Forest Series utilizes an innovative modular architecture, featuring 4 MB of Unified L2 cache with a latency of 17 cycles per four-core cluster, culminating in a total of 288 MB of L2 cache. This setup promises significantly improved bandwidth, with capacity increasing up to 400 GB/s—double the previous generation. The measured IPC uplift stands at 17%, based on benchmarks from SpecIntRate’17.

Clearwater Forest E-core module architecture with Enhanced L2 cache.

3D Chiplet Design and Connectivity

Intel’s Clearwater Forest integrates a total of 12 CPU chiplets fabricated with the 18A process, situated on three individual base tiles that encompass essential components such as the Fabric, LLC, memory controllers, and I/O, grounded on the Intel 3 process node. In addition, two I/O chiplets built on the Intel 7 node provide high-speed interconnects.

Diagram detailing Intel's 18A chip architecture with CPU, base, and I/O chiplets.

Summary of Clearwater Forest Configuration

The overall architecture encompasses:

12 E-Core CPU Chiplets (manufactured with Intel 18A)
3 Base Tile Packages (fabricated on Intel 3)
2 I/O Chiplets (engineered on Intel 7)

Chiplet architecture illustrating Intel's innovative design with enhanced efficiency.

Enhanced Performance Capabilities

Intel has emphasized the performance capabilities of a dual-socket Clearwater E-Core Xeon solution, which supports up to 3 TB of 12-channel DDR5-8000 memory, delivering a remarkable 1300 GB/s memory bandwidth. This platform accommodates 2 x 96 PCIe Gen 5 and 64 CXL lanes, as well as 144 UPI lanes (576 GB/s), achieving a grand total of up to 59 TF/s with 576 cores and 1152 MB of LLC, resulting in an impressive raw bandwidth of 5000 GB/s.

Source & Images