How does Xeon 6+ compare to AMD EPYC?

Intel claims Xeon 6+ offers approximately 30% better performance per thread than AMD EPYC 9965 in certain benchmarks. The 288 E-core design gives it a density advantage, and Intel specifically positions it for agentic AI workloads where CPUs handle orchestration (tools, files, rules, steps) around AI agents. A single liquid-cooled rack with Xeon 6+ delivers 36,864 cores in 32U of space, capable of running up to 150,000 agents.

Is Xeon 6+ good for AI workloads?

Yes, but for specific parts of the AI stack. Xeon 6+ is positioned as the CPU for 'agentic AI' — handling orchestration, tool calling, rule execution, and pipeline management around AI models. It's not designed to replace GPUs for training or inference of large models. Intel demonstrated agentic AI workloads at Computex where Xeon 6+ handles the orchestration layer while GPUs handle model inference.

When is Xeon 6+ available?

Xeon 6+ systems are available immediately through partners including ASUS, Dell, Ericsson, GIGABYTE, HPE, Lenovo, and Supermicro. The processors are shipping to enterprise server vendors as of Computex 2026.

Quick Answer

What Is Intel Xeon 6+ (Clearwater Forest)? 288-Core AI CPU

Q: What is Intel Xeon 6+ Clearwater Forest?

Intel Xeon 6+ (codename Clearwater Forest) is Intel's newest data center processor, announced at Computex 2026 on June 2, 2026. It's an all-Efficient-core design with up to 288 Darkmont E-cores per socket (576 cores dual-socket), built on Intel's 18A process node. It features 576MB of L3 cache, 12-channel DDR5-8000 memory, and consumes 300-450W TDP. Intel claims 48% better performance per watt over previous Xeon 6 E-core processors.

Published: June 3, 2026

What Is Intel Xeon 6+ (Clearwater Forest)? 288-Core AI CPU

Intel launched Xeon 6+ (Clearwater Forest) at Computex 2026, its first data center CPU built on the Intel 18A process node. With 288 E-cores per socket, it’s designed for the agentic AI era.

Last verified: June 3, 2026

Quick facts

Property	Value
Codename	Clearwater Forest
Release	June 2, 2026 (Computex 2026)
Process	Intel 18A
Cores	Up to 288 Darkmont E-cores per socket
Dual-socket	576 cores
L3 cache	576MB
L2 cache	288MB
Memory	12-channel DDR5-8000
PCIe	96 lanes Gen 5
TDP	300–450W
Design	All-E-core chiplet (12 chiplets on 18A)
Packaging	Foveros Direct 3D + EMIB

Architecture deep dive

Xeon 6+ is a radical departure from traditional Xeon designs:

All-E-core, no P-cores. Intel’s bet is that Efficient cores with massive cache and high bandwidth can outperform traditional Performance+Efficiency hybrid designs for scale-out workloads — especially agentic AI orchestration.

Chiplet design — 12 CPU chiplets on Intel 18A, three base tiles on Intel 3, and two I/O tiles on Intel 7, all connected via Foveros Direct 3D and EMIB packaging.

Massive cache — 576MB L3 + 288MB L2 = 864MB total cache per socket. This is critical for AI agent workloads where model weights and context data need fast access.

AI positioning: “The agentic AI CPU”

Intel is positioning Xeon 6+ specifically for agentic AI orchestration, not GPU-like training. The reasoning:

AI agents need CPUs for tool calling, rule execution, file system operations, and pipeline orchestration
Modern agentic workloads are CPU-bound at the orchestration layer, not GPU-bound
Xeon 6+ can handle 150,000 agents per rack (36,864 cores in 32U liquid-cooled)

Workload	Best processor
AI training	GPU (NVIDIA, AMD)
Model inference	GPU / AI accelerator
Agent orchestration	Xeon 6+ CPU
Tool execution	Xeon 6+ CPU
Data preprocessing	Xeon 6+ CPU

Performance claims

Intel’s published benchmarks vs AMD EPYC 9965:

Metric	Intel Xeon 6+	vs AMD EPYC 9965
Performance per watt	+48% vs prev gen	Not directly compared
Performance per thread	~30% faster	Leading claim
Rack density	36,864 cores/32U	Competitive
Agent capacity per rack	150,000 agents	Not specified

Rack-scale infrastructure

Intel partnered with Foxconn, SambaNova, and Vista Equity for rackscale AI:

Foxconn — rack-scale product development
SambaNova — AI inference solutions (SN50 SambaRack with Xeon 6)
Vista Equity — enterprise financing and deployment

A single liquid-cooled rack: 36,864 cores, 32U of compute, capable of 150,000 concurrent agents.

Availability

Xeon 6+ systems are available immediately from: ASUS, Dell, Ericsson, GIGABYTE, HPE, Lenovo, Supermicro

Intel CEO Lip-Bu Tan delivered the Computex 2026 keynote, emphasizing execution, engineering, and ecosystem partnerships.

Bottom line

Xeon 6+ is Intel’s most ambitious data center CPU play in years. Its focus on agentic AI orchestration — rather than trying to beat NVIDIA at GPU workloads — is a smart bet on where the industry is heading. The 288-core, 18A, 576MB cache design is genuinely impressive. For enterprise AI infrastructure planning, Xeon 6+ deserves serious consideration for the orchestration layer of agentic systems.