Intel’s 288-Core Clearwater Forest Xeon 6+ Lands on 18A, Claiming 30% Performance & 50% Efficiency Lead Over AMD’s 192-Core EPYC

Intel has finally launched its next-gen Xeon 6+ E-core CPU family codenamed Clearwater Forest, which packs 288 cores & is built on 18A.

Intel Clearwater Forest "Xeon 6+" CPUs Begin Rolling Out With 288 E-Cores - The First Data Center Product Built on 18A

Intel brought higher compute density and performance per watt gains with its last E-Core-only Xeon CPU offering, Sierra Forest. This was also the first time that Intel segmented its Xeon lineup into P-Core and E-Core only families.

Now, the journey continues with the 2nd Gen E-Core-only family codenamed Clearwater Forest, which will be part of the Xeon 6+ lineup.

2 of 9

It Has It All - Intel 18A With RibbonFET & Power Via, Forveros Direct3D, EMIB 2.5D

With Clearwater Forest, Intel is taking its disaggregated architecture and packaging design to the next level.

The chip is a multi-layered solution with several chiplets and building blocks that make it quite an engineering achievement from Intel.

One Clearwater Forest or Xeon 6+ CPU is composed of twelve EMIB tiles that use 2.5D packaging. These tiles bring three active base tiles together, which are then connected to two I/O Tiles and a total of twelve compute Tiles. The I/O Tiles utilize the Intel 7 process node, the Active Base Tiles utilize the Intel 3 process node, and the compute chiplets are fabricated on the Intel 18A process node.

2 of 9

Coming to the compute chiplets, these Darkmont E-Core chiplets are fabricated using the 18A process node technology with RibbonFET. Intel claims that 18A delivers the best power efficiency for core logic through lower gate capacitance.

18A also offers higher cell density and over 90% cell utilization, along with improved signal routing through power rails on the backside. 18A also enables low-loss power delivery, reducing losses by 4-5%.

Diving into the RibbonFET technology of 18A, this allows greater electrical current control and reduced power leakage for a significant performance advantage. With RibbonFET, the gate surrounds the transistor channel, giving really tight control over the electrical currents in the channel, allowing more drive and less leakage. This also contributes to lower voltage operating levels. Also, RibbonFET's gate length is 5-10% shorter than FinFET's, and it also achieves 20% power reduction per transistor.

Some features of RibbonFET include:

• Further miniaturization of chip components is critical in high-density CPUs
• Tight control over the electrical current in the transistor channel
• Improves perf/watt, (Vmin) operations, and electrostatics
• Tunability through ribbon widths and multiple threshold voltage types

Then there's PowerVia, which improves standard cell utilization by up to 10% and ISO-power performance by up to 4%. PowerVia allows Intel to drive power from underneath the silicon, or Backside power in a way.

Some highlights of PowerVia include:

• Reduced congestion and boosted performance
• Relocates course pitch metals
• Bumps to the back side of the die
• Nano-Scale TSVs for efficient power distribution
• Improved Signal Routing
• Higher cell density - over 90% cell utilization

Clearwater Forest is also the first high-volume production CPU that leverages Foveros Direct3D tech, which is an advanced packaging solution that bridges the compute and IO tiles together on the base active tiles. Foveros Direct 3D has a 9 μm bump pitch and uses Cu-to-Cu bonding. It acts as an active Si interposer with high density and low resistance and offers ~0.05pJ/bit performance. This means that Intel needs to spend virtually zero power to move data through and forth between the two dies.

The following is a 3D Construction overview of the Clearwater Forest Xeon 6+ CPU:

Dissecting The Three Main Tiles of Clearwater Forest

Alright, so next up, we get to see what each of the three main tiles has to offer. Once again, there are three tiles: the Compute Tile, the I/O Tile, and the Base Tile.

Clearwater Forest I/O Tile

The Clearwater Forest IO Tile is fabricated on the Intel 7 process technology. This tile features a total of eight accelerators in two packages, which offer Intel Quick Assist Technology, Intel Dynamic Load Balancer, Intel Data Streaming Accelerator, and Intel In-Memory Analytics Accelerator. There are 16 accelerators across the two I/O tiles.

Each I/O Tile also offers 48 PCIe Gen 5.0 lanes (combined total of 96), 32 CXL 2.0 lanes (combined total of 64), and 96 UPI 2.0 lanes (combined total of 192). The IO Tile remains unchanged from Granite Rapids, but is a clear upgrade over Sierra Forest.

Clearwater Forest Base Tile

Moving on to the Base Tile, this Tile is connected using EMIB, which connects to the compute Tiles above it. There are three Base Tiles fabricated on the Intel 3 process technology. Each Base Tile carries four DDR5 memory controllers for a total of 12 memory channels on the chip. The Tile also packs a shared LLC with 48 MB per compute tile or 192 MB per base tile. This provides a massive 576 MB of on-package LLC.

Clearwater Forest Compute Tile

The compute tiles on Clearwater Forest are probably the most interesting ones on the chip since they are based on the new 18A process technology. Each compute tile is composed of 6 modules, & each module packs 4 Darkmont E-Cores. That gives us 24 Darkmont E-Cores per compute tile and 288 E-Cores across 12 compute tiles.

2 of 9

Each module also packs 4 MB of L2 cache, which means that you are looking at 24 MB of L2 cache per tile and 288 MB of total L2 cache across 12 compute tiles. This is the same as the Sierra Forest E-Core CPUs & gives us a combined L3+L2 cache of 864 MB across the entire chip.

So you have the following:

• 12x Compute Tiles (Intel 18A)
• 3x Active Base Tiles (Intel 3)
• 2x Intel I/O Tiles (Intel 7)
• 12x EMIB Tiles (EMIB 2.5D)

Darkmont E-Core Deep-Dive

Now we get to talk about the Darkmont E-Core, which has also been used in Panther Lake client CPUs.

This core is largely similar to the Skymont architecture we saw on Lunar Lake and Arrow Lake CPUs, but, compared to Crestmont, it is a big upgrade.

Starting with the details, Skymont comes with an updated prediction block with 128 bytes, faster "Find the next" instructions, and 96 instruction bytes for Parallel Fetch. Darkmont is a 9-wide microarchitecture and also features a wider Decode, which includes 9-wide (3x3) or 50% more decode clusters than Crestmont E-Cores, a Nanocode that unlocks microcode parallelism per cluster, and a Uop queue capacity that's increased from 64 entries to 96 entries. There's also a larger 64KB Instruction Cache alongside accurate and enhanced branch prediction.

On the front-end side (OOE or Out of Order Engine), we are looking at an 8-wide allocation and 16-wide retire, which means that resources can be added and cleared faster. Queuing also gets more resources, with the out-of-order window now growing to 416 entries.

Dispatch ports have been increased to 26. The Scalar Engine gets 8 Integer ALUs, 3 load & 4 store AGU ports, 3 jump ports, 2 integer store data ports, & the Vector Engine gets 4 vector/float ALUs, 2 vector/float store data, 4 vector/float stacks.

The memory subsystem enhancements see an increase across the board, with the L2 cache being 4 MB L2 per four-core clusters, double the bandwidth from 64B to 128B/ cycle, and L1 to L1 transfers are now faster and offer a more predictable communication.

This is achieved by eliminating the need to transfer data from the fabric and instead just going to the L2 cache via the L1 cache. The conviction clock has been upgraded from 16 bytes to 32 bytes per clock.

The following is the comparison between Crestmont and Darkmont E-Core architectures:

Bringing it all together, Darkmont E-Cores on Clearwater Forest offer up to 90% higher performance than the 144-core Xeon 6780E "Sierra Forest" CPU, offering 23% improvement in efficiency across the load line & up to 8:1 server consolidation with lower TCO.

Performance Benchmarks

Intel has also shared a few performance metrics for Clearwater Forest "Xeon 6+" CPUs. The comparison includes the Xeon 6780E "Sierra Forest" chips with 144 cores, and the AMD EPYC 9965 "Turin" with 192 cores against the Xeon 6990E+ "Sierra Forest" chips with 288 cores.

Compared to the 144-core Sierra Forest (Xeon 6780E) at 330W, the Clearwater Forest chip with 288 cores and 450W TDP offers a 36.3% lower TDP, twice as many cores, and 2.26x higher average performance, & 55% higher average performance/watt versus its predecessor.

Compared to the AMD EPYC 9965, which is a 192-core chip based on the Zen 5C architecture, the Xeon 6+ 288-core offers a 30% uplift in performance and efficiency on average across a range of workloads at 450W, while the EPYC chip has a TDP of 500W.

For Security and Crypto, the onboard controller delivers a massive 15.2x leap against the Xeon 6780E & a 6.2x leap against the AMD EPYC in SMx. For SHA, the Xeon 6+ offers a 5.2x gain over the Xeon 6780E & a 2.6x gain over the EPYC 9965.

Intel also shares a typical server utilization scenario where the Xeon 6990E+ is compared against the EPYC 9965 at a typical 40% CPU utilization. Here, the Xeon 6+ achieves a 30% higher performance efficiency vs the EPYC CPU.

The performance uplift is achieved with the new Darkmont E-Cores, which have a 17% uplift in IPC. With these, Clearwater Forest brings 1.9x higher performance, 23% improvement in efficiency, and up to 9:1 server consolidation versus an aging Xeon platform.

The only thing to note here is that AMD is already mass-producing its next-gen EPYC Venice platform with up to 256 of its latest Zen 6 cores. So this lead (if there remains any) will shrink based on the performance disclosures that AMD has made so far.

Intel Xeon 6+ CPU & Platform Details

Now for the platform details, Intel's Clearwater Forest "Xeon 6+ CPUs will be supported on the LGA 7529 socket in 1S & 2S configurations. This is the same socket used by the Xeon 6900P "Granite Rapids-AP" CPUs. The same socket was also going to use the 288-core version of Sierra Forest "Xeon 6900E," though those were cancelled. The chips will be rated at 300-500W TDP, which is the same operating range as the Xeon 6700E and 6900P CPUs. The lower TDP spec will also come with half the number of cores, such as 144, the same as Xeon 6700E.

The chips will be able to support up to 12-channel DDR5 memory with speeds of up to 8000 MT/s. In addition to that, the platform will support up to 6 UPI 2.0 links (up to 24 GT/s per lane), up to 96 PCIe Gen5.0 lanes (x16,x8,x4,x2), and up to 64 CXL 2.0 lanes.

Security features will include Intel Software Guard Extensions or SGX and Intel Trust Domain Extensions or TDX. On the power management side, these chips will carry Intel AET (Application Energy Telemetry) & Intel Turbo Rate Limiter. And lastly, Clearwater Forest CPUs will get Advanced Vector Extensions 2 with VNNI and INT8 support.

2 of 9

The lineup will include four main chips, with the two 200+ core-count variants offered at two different TDP levels. The Xeon 6990E+ will be the flagship with 288 cores, up to 3.2 GHz boost clocks, 576 MB of cache, 96 PCIe lanes, and a 450W TDP. The 330W variant of this chip will feature a lower boost and all-core boost of 1.7 GHz/2.4 GHz versus 2.2/2.8 GHz on the 450W part.

There's also the Xeon 6980E+ with 264 cores, 528 MB of L3 cache, 96 PCIe lanes, and a 400W TDP, which operates at 2.1 GHz base, 2.7 GHz all-core boost, and 3.2 GHz boost clocks, while the 300W variant of this chip operates at 1.6 GHz base and 2.2 GHz all-core boost clocks.

The sub-200 core variants include the Xeon 6970E with 192 cores, 480 MB of L3 cache, a 400W TDP, and the Xeon 6960E+ with 144 cores, 432 MB of L3 cache, and a 330W TDP.

So, rounding up Intel's Clearwater Forest "Xeon 6+" versus Sierra Forest "Xeon 6", we will see:

• Up To 2x Core Count
• 17% IPC per core
• >5x Last Level Cache
• +4 Memory Channels
• +2 UPI Links
• 20% Faster memory speed

Intel Xeon 6+ CPUs will be available in various 1S and 2S options from leading OEMs as of today, fueling the Agentic AI, 5G, and Telcom segments with higher density & performance.

2 of 9

Intel Xeon CPU Families (Preliminary):

Family Branding	Coral Rapids	Diamond Rapids	Clearwater Forest	Granite Rapids	Sierra Forest	Emerald Rapids	Sapphire Rapids	Ice Lake-SP	Cooper Lake-SP	Cascade Lake-SP/AP	Skylake-SP
Process Node	Intel 14A?	Intel 18A-P	Intel 18A	Intel 3	Intel 3	Intel 7	Intel 7	10nm+	14nm++	14nm++	14nm+
Platform Name	TBD	Intel Oak Stream	Intel Birch Stream	Intel Birch Stream	Intel Mountain Stream Intel Birch Stream	Intel Eagle Stream	Intel Eagle Stream	Intel Whitley	Intel Cedar Island	Intel Purley	Intel Purley
Core Architecture	TBD	Panther Cove-X	Darkmont	Redwood Cove	Sierra Glen	Raptor Cove	Golden Cove	Sunny Cove	Cascade Lake	Cascade Lake	Skylake
MCP (Multi-Chip Package) SKUs	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	No	Yes	No
Socket	TBD	LGA XXXX / 9324	LGA 4710 / 7529	LGA 4710 / 7529	LGA 4710 / 7529	LGA 4677	LGA 4677	LGA 4189	LGA 4189	LGA 3647	LGA 3647
Max Core Count	TBD	Up To 192 P-Cores	Up To 288	Up To 128	Up To 288	Up To 64?	Up To 56	Up To 40	Up To 28	Up To 28	Up To 28
Max Thread Count	TBD	Up To 192	Up To 288	Up To 256	Up To 288	Up To 128	Up To 112	Up To 80	Up To 56	Up To 56	Up To 56
Max L3 Cache	TBD	TBD	TBD	480 MB L3	108 MB L3	320 MB L3	105 MB L3	60 MB L3	38.5 MB L3	38.5 MB L3	38.5 MB L3
Memory Support	TBD	Up To 16-Channel DDR5-9000+	Up To 12-Channel DDR5-8000	Up To 12-Channel DDR5-6400 MCR-8800	Up To 12-Channel DDR5-6400	Up To 8-Channel DDR5-5600	Up To 8-Channel DDR5-4800	Up To 8-Channel DDR4-3200	Up To 6-Channel DDR4-3200	DDR4-2933 6-Channel	DDR4-2666 6-Channel
PCIe Gen Support	PCIe 6.0	PCIe 6.0	PCIe 5.0 (96 Lanes)	PCIe 5.0 (136 Lanes)	PCIe 5.0 (88Lanes)	PCIe 5.0 (80 Lanes)	PCIe 5.0 (80 lanes)	PCIe 4.0 (64 Lanes)	PCIe 3.0 (48 Lanes)	PCIe 3.0 (48 Lanes)	PCIe 3.0 (48 Lanes)
TDP Range (PL1)	TBD	TBD	Up To 500W	Up To 500W	Up To 350W	Up To 350W	Up To 350W	105-270W	150W-250W	165W-205W	140W-205W
3D Xpoint Optane DIMM	TBD	TBD	N/A	Donahue Pass	N/A	Crow Pass	Crow Pass	Barlow Pass	Barlow Pass	Apache Pass	N/A
Competition	TBD	AMD EPYC Venice	AMD EPYC Turin	AMD EPYC Turin	AMD EPYC Bergamo	AMD EPYC Genoa ~5nm	AMD EPYC Genoa ~5nm	AMD EPYC Milan 7nm+	AMD EPYC Rome 7nm	AMD EPYC Rome 7nm	AMD EPYC Naples 14nm
Launch	2028-2029	2027	2026	2024	2024	2023	2022	2021	2020	2018	2017

Follow Wccftech on Google to get more of our news coverage in your feeds.