Intel Launches Xeon 6+ 'Clearwater Forest' on 18A, Its First Data Center CPU on the New Node, With Up to 288 Cores
Intel's first 18A server chip ships with up to 288 Darkmont E-cores and 576 MB of L3 cache, claiming a 30% per-thread lead over AMD's 192-core EPYC 9965.
Overview
Intel launched its Xeon 6+ server processors, code-named Clearwater Forest, at Computex 2026, marking the first data center CPU built on the company’s 18A manufacturing process, according to The Register. Intel describes Clearwater Forest as its first 18A-based server processor, as noted in an Intel Newsroom preview ahead of the launch. The release puts Intel’s most advanced node into volume server silicon as the company works to defend its data center position against AMD and an expanding field of Arm-based competitors.
What We Know
Clearwater Forest is an all-efficiency-core design. The lineup tops out at 288 of Intel’s Darkmont E-cores, according to The Register and TechTimes. The flagship Xeon 6990E+ carries 576 MB of L3 cache, with the cache scaling down to 432 MB on the 144-core entry model, the Xeon 6960E+, as reported by The Register.
The chip is assembled from tiles spanning multiple Intel processes. According to The Register, the package combines twelve 24-core compute tiles built on Intel 18A, three base tiles on Intel 3 that hold the memory controllers and L3 cache, and two I/O dies carried over from the earlier Xeon 6900P series. Intel’s Kira Boyko, Product Line Director for E-Core Xeon Products, characterized the design in a roundtable as “our most performant Xeon on the market today, specifically for scale-out workloads,” per a transcript published by Tom’s Hardware.
On the data sheet, all Clearwater Forest parts reach a 3.2 GHz maximum turbo frequency, span a 330-to-450-watt TDP range, and support 12 channels of DDR5 memory at 8000 MT/s, according to The Register. Intel says the new E-cores deliver 17 percent higher instructions per clock than those in its first generation of E-core Xeons, as reported by The Register and stated in the Intel Newsroom preview.
For competitive positioning, Intel says the 6990E+ delivers 30 percent higher average performance per thread than AMD’s 192-core EPYC 9965, according to TechTimes. Against its own prior-generation flagship, the Xeon 6780E, Intel claims 55 percent better performance per watt, also per TechTimes.
The flagship Xeon 6990E+ is available immediately through Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro, according to TechTimes. The Register notes the processors offer drop-in compatibility with existing boards, easing OEM adoption.
The launch also underscores Intel’s bet on E-core-heavy designs for cloud and scale-out work. Tim Wilson, Vice President and General Manager of Intel’s Data Center Silicon Engineering group, framed the market opportunity by saying that “data centers that have built on GPUs for the last three years are suddenly finding they’re bottlenecked by the CPU,” as quoted by TechTimes. On the decision to forgo simultaneous multithreading in the E-core line, Wilson told the Tom’s Hardware roundtable that “one or two physical cores are always better than two virtual cores built on one physical core.”
Kevork Kechichian, Executive Vice President and General Manager of Intel’s Data Center Group, said Clearwater Forest “empowers data centers, cloud providers and telecom operators to scale workloads more efficiently, dramatically reduce energy consumption, and deliver smarter, more responsive services,” according to the Intel Newsroom.
What We Don’t Know
Intel’s headline comparison figures are vendor-supplied and have not yet been independently benchmarked. The 30 percent per-thread claim is measured against AMD’s EPYC 9965, while the 55 percent performance-per-watt figure is relative to Intel’s own previous-generation Xeon 6780E, not to AMD; the two numbers describe different baselines and should not be read as a single combined advantage. How Clearwater Forest performs in third-party testing, and how AMD’s competing parts respond, remains to be seen.
The ramp of Intel’s 18A node also remains a constraint. In the launch roundtable, Kira Boyko said wafer allocation decisions are being made “daily, in some cases,” and Wilson said Intel gives “as many CPUs to as many people as we can,” describing the calls as “more business decisions than engineering decisions,” according to Tom’s Hardware. That language points to supply being actively managed rather than abundant at launch.
Analysis
Clearwater Forest is significant less for any single specification than for what it represents: high-volume Intel server silicon shipping on 18A, the node on which the company has staked much of its turnaround. The all-E-core approach mirrors AMD’s dense-core EPYC strategy and the rise of Arm designs in hyperscale fleets, where core count and performance-per-watt for scale-out workloads increasingly matter more than peak single-thread speed. By leaning on Darkmont E-cores and a multi-tile package that mixes 18A, Intel 3, and reused I/O dies, Intel is signaling that its chiplet manufacturing strategy is now mature enough for its flagship data center product, even as its own engineers acknowledge that 18A supply is being rationed in the near term.