Quick Verdict
An extremely dense E‑core server CPU that trades per‑core speed for sheer throughput and efficiency, ideal for cloud and telecom deployments that can exploit its 288 cores and 12‑channel DDR5‑8000 memory.
Overview
Launch
2026
Status
LaunchedGeneration
Xeon 6+ (Clearwater Forest)
Market
Hyperscale / Cloud / Telecom / Scale‑out Data Center
The Intel Xeon 6990E+ is a 288‑core E‑core server processor built on Intel 18A, designed for dense, throughput‑oriented cloud, telecom, and scale‑out workloads where performance per watt and core count matter more than peak per‑core performance.
Built on Intel’s 18A process using Darkmont E‑cores in a chiplet design, the Xeon 6990E+ packs 288 cores, 576 MB of L3 and 288 MB of L2 cache, and 12‑channel DDR5‑8000 into a 450 W TDP envelope. It targets hyperscale data centers, telcos, and AI inference hosts that need many threads and high memory bandwidth more than ultra‑low latency or high per‑core performance. Intel claims up to 30% better performance per thread versus AMD’s 192‑core EPYC 9965 and around 2.
3× the performance of the previous 144‑core Xeon 6780E, but real‑world advantage will depend on your actual workload and scaling.
Specifications
Performance
Exceptional for highly threaded server workloads that scale to many cores; limited benefit for lightly threaded office or creator apps.
Excellent for dense VM and container hosting, with high core counts, large memory, and strong I/O.
Not a gaming CPU; low per‑core clocks and E‑core design make it uncompetitive for latency‑sensitive gaming workloads.
Intel claims substantial performance‑per‑watt gains vs prior Xeon E‑core parts; real‑world efficiency will depend on workload and power management.
- •Not intended for gaming; low clock speeds and E‑core architecture.
- •Latency and single‑thread performance far below mainstream desktop CPUs.
- •Only relevant if gaming is a tiny side‑load on a primarily cloud server.
- •Strong CPU inference throughput for models that fit in memory.
- •No integrated GPU or dedicated AI accelerator; relies on CPU cores and Intel DL Boost / AVX2.
- •Best used as a host CPU for GPU/accelerator‑based AI systems.
Architecture
Intel 18A
Process Node
Clearwater Forest
Codename
288C / 288T
Core Config
576 MB
L3 Cache
450 W
TDP
Architecture Overview
Xeon 6990E+ is a multi‑chiplet processor using Intel 18A compute tiles stacked on Intel 3 base tiles and Intel 7 I/O tiles, connected via EMIB and Foveros Direct 3D packaging. It continues the ‘Forest’ E‑core line but moves from Crestmont (Sierra Forest) to Darkmont cores on 18A, with a large distributed cache and 12‑channel DDR5 memory subsystem.
CPU Design
288 Darkmont E‑cores arranged in 12 compute tiles of 24 cores each, without SMT; four cores share a 4 MB unified L2 cache, yielding 288 MB L2 in total, plus up to 576 MB of L3 cache distributed across the base tiles.
Memory Subsystem
12‑channel DDR5‑8000 RDIMM interface supports up to 1.5 TB of memory per socket, with high aggregate bandwidth for data‑center workloads. Intel 18A’s PowerVia backside power delivery helps supply clean power to the many cores and memory channels.
PCIe & I/O
96 PCIe 5.0 lanes from the CPU, with 64 lanes configurable as CXL for cache‑coherent accelerator access. Six UPI 2.0 links at 24 GT/s enable dual‑socket interconnects with high coherency bandwidth.
Overclocking
Multiplier is locked; tuning focuses on power modes (SST‑PP) rather than traditional overclocking. Intel SST‑PP allows different base/TDP points, for example a 1.7 GHz / 330 W mode for denser, lower‑power deployments.
- Double the cores (144 → 288) and threads (144 → 288).
- Move from Crestmont E‑cores to Darkmont E‑cores with ~17% IPC uplift.
- Intel 18A process instead of Intel 7, enabling higher density and better performance per watt.
- 576 MB L3 + 288 MB L2 vs 108 MB total cache on 6780E.
- DDR5‑8000 support vs DDR5‑6400 on Sierra Forest, with 12 channels retained.
Key Highlights
- 288 E‑cores for extremely high thread density.
- Intel 18A process with improved performance per watt and density.
- 576 MB L3 + 288 MB L2 cache reduces memory bottlenecks.
- 12‑channel DDR5‑8000 with up to 1.5 TB capacity per socket.
- 96 PCIe 5.0 / CXL lanes for accelerators and NVMe.
- Strong Intel claims vs prior Xeon E‑core and competing EPYC density parts.
- No SMT; 288 threads are fixed, no 2× SMT multiplier.
- E‑core clocks and single‑thread performance are modest.
- 450 W TDP requires robust cooling and power delivery.
- Platform and CPU cost will be very high; overkill for small deployments.
- New 18A node and complex packaging may introduce early‑production risk.
History
The Xeon 6990E+ is the flagship of Intel’s Clearwater Forest generation, which itself is the successor to Sierra Forest (Xeon 6700E series). Sierra Forest introduced the concept of an E‑core‑only Xeon line for cloud density, but topped out at 144 Crestmont cores. Clearwater Forest moves to Intel 18A and Darkmont E‑cores, doubles the maximum core count to 288, and significantly expands cache and memory bandwidth.
It launched at Computex 2026 as Intel’s first 18A‑based data‑center CPU, coming at a time when Intel is trying to re‑assert leadership in density‑optimized servers against AMD’s high‑core‑count EPYC parts and Arm‑based alternatives like AmpereOne. The 6990E+ is effectively Intel’s answer to the question: what if you could pack 288 efficient cores into a single socket and still run them within a credible power envelope?
Improvements over Previous Generation
- Double the cores (144 → 288) and threads (144 → 288).
- Move from Crestmont E‑cores to Darkmont E‑cores with ~17% IPC uplift.
- Intel 18A process instead of Intel 7, enabling higher density and better performance per watt.
- 576 MB L3 + 288 MB L2 vs 108 MB total cache on 6780E.
- DDR5‑8000 support vs DDR5‑6400 on Sierra Forest, with 12 channels retained.
Alternatives & Competitors
Should You Buy It?
Recommended for the right buyer
New cloud / telecom deployments that can fully utilize 288+ cores and 12‑channel DDR5‑8000, and where performance per watt and rack density are more important than per‑core speed.
Avoid if…
- You need high per‑core performance (databases with complex queries, latency‑sensitive trading).
- Your workloads are lightly threaded and cannot scale beyond ~64 cores.
- You want a simple workstation or gaming CPU.
- Power and cooling constraints are severe and you cannot accommodate ~450 W per socket.
Use Cases
Interesting Facts
First data‑center CPU manufactured on Intel’s 18A node in high volume, using RibbonFET and PowerVia.
Uses a chiplet design with 12 compute tiles, three base tiles, and two I/O tiles connected via EMIB and Foveros Direct 3D stacking.
Darkmont E‑cores deliver around 17% IPC uplift over the Crestmont cores used in Sierra Forest, according to Intel.
A dual‑socket 6990E+ system can reach 576 physical cores and 3 TB of DDR5 memory.
Intel positions it against AMD’s 192‑core EPYC 9965, claiming up to 1.3× higher performance per thread and per thread per watt in selected benchmarks.
SST‑PP allows operators to trade base clock for TDP, enabling 1.7 GHz / 330 W modes for lower‑power racks.
Large 576 MB L3 cache is split into 8 MB slices per quad‑core cluster, optimizing for throughput workloads.
Designed primarily for cloud, telco, and AI inference rather than traditional HPC or workstations.
Supports Intel AET (Application Energy Telemetry) for real‑time power and performance monitoring at the application level.
Clearwater Forest is the second ‘Forest’ E‑core generation, following Sierra Forest (Xeon 6700E series).
People Also Ask
Is the Intel Xeon 6990E+ good for gaming?
No. It has many cores but low per‑core clocks and E‑core architecture, making it unsuitable for gaming compared to mainstream desktop CPUs.
How much memory does the Xeon 6990E+ support?
Up to 1.5 TB of DDR5‑8000 RDIMM memory across 12 channels.
Does the Xeon 6990E+ have Hyper‑Threading?
No. Intel’s E‑core Xeon 6+ parts do not use SMT/Hyper‑Threading; each core is single‑threaded.
What socket does the Xeon 6990E+ use?
LGA7529, the same socket used by Xeon 6900P/6900E Granite Rapids‑AP and Sierra Forest‑AP parts.
How does Xeon 6990E+ compare to EPYC 9965?
Intel claims up to 1.3× higher performance per thread and per thread per watt vs EPYC 9965, but EPYC uses SMT and fewer cores; real‑world leadership will vary by workload and needs independent validation.
Is the Xeon 6990E+ overclockable?
Not in the traditional sense; the multiplier is locked, but Intel SST‑PP allows configurable power and frequency profiles.
What process node is the Xeon 6990E+ built on?
Intel 18A, Intel’s 2 nm‑class node with RibbonFET transistors and PowerVia backside power delivery.
Does the Xeon 6990E+ have integrated graphics?
No, it has no integrated GPU; display output requires an add‑in GPU or BMC, which is typical for server platforms.
How many PCIe lanes does the Xeon 6990E+ provide?
96 PCIe 5.0 lanes from the CPU, with 64 configurable as CXL lanes for accelerators.
What is the TDP of the Xeon 6990E+?
450 W base TDP; Intel SST‑PP also offers a 330 W mode with reduced base frequency.
Frequently Asked Questions
Can I use the Xeon 6990E+ in a workstation?
Technically yes on a compatible LGA7529 workstation board, but it’s optimized for server workloads. Most workstation users will be better served by P‑core Xeon 6 or Xeon W parts.
Does the Xeon 6990E+ support AVX‑512?
Public Intel Ark listings for Xeon 6+ 6990E+ show AVX2 but not AVX‑512. AVX‑512 support is not confirmed for this SKU at launch; check the latest Intel datasheet if your workload depends on it.
Is the Xeon 6990E+ suitable for virtualization?
Yes. With 288 cores, 12 memory channels, and support for Intel VT‑x, VT‑d, and large memory, it is excellent for dense VM and container environments.
What cooling does the Xeon 6990E+ require?
450 W TDP typically requires high‑end air or liquid cooling in a well‑designed server chassis; follow your system vendor’s guidance.
Can I run AI inference on the Xeon 6990E+ without GPUs?
You can, but performance will be limited compared to GPU or dedicated accelerators. It’s better suited as a host CPU for GPU‑based inference systems.
Does the Xeon 6990E+ support Intel AMX?
Intel has not publicly listed AMX support for Xeon 6+ E‑core SKUs like the 6990E+ in Ark materials. If AMX is required, verify support in the official Intel datasheet or consider P‑core Xeon 6 parts where AMX is confirmed.
Is the Xeon 6990E+ a good choice for databases?
Good for scale‑out, in‑memory, or analytics databases that can use many threads; less ideal for latency‑sensitive OLTP workloads that benefit from higher per‑core performance.
What chipsets support the Xeon 6990E+?
It uses the Intel Xeon 6+ server platform with LGA7529; refer to Intel’s server platform documentation and vendor boards for chipset details.
How does the Xeon 6990E+ differ from the Xeon 6980E+?
The 6990E+ has 288 cores and 576 MB L3 at 450 W; the 6980E+ has 264 cores, 528 MB L3, and 400 W TDP. Both are E‑core Clearwater Forest parts on the same platform.
Is the Xeon 6990E+ power‑efficient?
Relative to prior Xeon E‑core parts, Intel claims significant performance‑per‑watt gains. However, absolute power is still high (450 W), so efficiency depends on how fully you utilize the cores.