Skip to content
TECH CEO Daily
AILaunch

Intel Diamond Rapids Xeon scales to 256 cores; Crescent Island inference GPU gets up to 480GB

Intel's next Xeon pairs 256 cores with 16 memory channels on its 18A-P process, while its Crescent Island inference card packs up to 480GB of memory in a 350-watt air-cooled design.

By · Editor

Added to archive · Covers events of August 24, 2026 · 3 min read · Fact-checked

The 60-second brief

  • 1Intel said on August 24, 2026, that Diamond Rapids Xeon scales to 256 cores and 1.28GB of cache on 18A-P.
  • 2Crescent Island, an inference GPU, carries up to 480GB of LPDDR5X memory in a 350-watt, air-cooled PCIe card.
  • 3Intel gave no launch dates and, per Chips and Cheese, withheld Crescent Island's memory bandwidth, so buyers should demand benchmarks.

The news

Intel Diamond Rapids, the company's next Xeon server processor, will scale to 256 cores, Intel said on August 24, 2026, at the Hot Chips conference. Intel (INTC) also detailed Crescent Island, an inference GPU with up to 480GB of memory, pitching both at agentic AI.

Agentic AI refers to software agents that plan and carry out multistep tasks, calling tools and other programs along the way. Intel described Diamond Rapids as a processor for enterprise-scale agentic AI that orchestrates workloads across cloud and enterprise systems.

According to Intel, Diamond Rapids offers up to 256 cores with 1.28GB of last-level cache, the on-chip memory shared by the cores. It supports 16 memory channels at 12,800 megatransfers per second, and 128 lanes of PCIe Gen6 and CXL 3.0, a standard for attaching extra memory and accelerators. It is built on Intel 18A-P, a performance-enhanced version of the company's 18A manufacturing process, and adds Advanced Performance Extensions (APX) and updated Advanced Matrix Extensions (AMX) for AI math.

Crescent Island is a data center GPU for inference, the work of running trained AI models. Intel said it combines 32 Xe3P cores and 256 XMX matrix engines with up to 480GB of LPDDR5X, a low-power memory type, on a 350-watt, air-cooled PCIe card. Intel said the design aims to deliver more tokens, the units of text AI models produce, for next-generation agentic workloads.

Intel also covered Wildcat Lake, sold as Intel Core Series 3, a six-core 18A chip with a neural processing unit rated at up to 17 trillion operations per second (TOPS) for laptops and edge devices. Intel said it is the first Intel processor to use UCIe, an open standard for connecting chiplets. Intel chief technology officer Pushkar Ranade said agentic AI is changing how Intel designs computing, "from the transistor and package up through the full system architecture."

Intel's post did not give launch dates for any of the chips or name the process for Crescent Island. Techzine reported on August 26 that Intel scheduled initial Crescent Island customer shipments for the second half of 2026. Chips and Cheese reported on August 27 that Intel declined to disclose the card's memory bandwidth when asked.

The numbers

Diamond Rapids maximum cores
256
Diamond Rapids last-level cache
1.28GB
Diamond Rapids memory channels
16 at 12,800 MT/s
Diamond Rapids I/O
128 lanes PCIe Gen6, CXL 3.0
Crescent Island memory
Up to 480GB LPDDR5X
Crescent Island card power
350 watts, air-cooled

Why CEOs should care

For infrastructure buyers, the specs set a planning baseline for the next round of server designs. Sixteen memory channels and 256 cores per processor point toward consolidating CPU-heavy agent orchestration onto fewer servers. Ask Intel and your server makers when Diamond Rapids platforms will ship, what a fully loaded processor draws, and how software licensed per core will be priced at that density.

Crescent Island is the more unusual bet. A 350-watt, air-cooled PCIe card can in principle go into existing server racks without liquid cooling, and StorageReview said the 480GB capacity lets the card host large models and long context windows. But Intel did not disclose memory bandwidth, which largely sets how fast a model can generate tokens. CFOs should hold capacity plans until independent tokens-per-second and tokens-per-watt results are available.

For boards and CIOs, the risk is timing. Diamond Rapids depends on Intel executing on its 18A-P manufacturing process, Intel has not said how Crescent Island will be made, and the company's post gave no ship dates for either. Treat these chips as an option to test against Nvidia and AMD systems, not a commitment.

The bigger picture

The disclosures fit Intel's argument that agentic AI will need far more general-purpose processors, not only GPUs. When Intel priced a $20 billion stock offering on August 11, Yahoo Finance reported that demand tied to AI agents had pushed CPU requirements beyond Intel's existing manufacturing capacity.

Crescent Island also gives Intel another entry in data center AI accelerators, where Chips and Cheese noted that Intel's GPU lineup has had a rocky history. That history is why the undisclosed details matter to buyers comparing it with Nvidia and AMD products.

What happened next

On August 26, 2026, Intel chief financial officer David Zinsner said at a Deutsche Bank conference that moving from AI training to agentic workloads raises CPU requirements roughly four to six times, according to an Investing.com transcript. On August 27, Chips and Cheese published an analysis noting Intel had said at OCP 2025 that Crescent Island would sample by the end of 2026, and wrote that the timeline may have slipped into 2027.

Watch for Crescent Island customer sampling or initial shipments before the end of 2026, a Diamond Rapids launch date, and independent performance results. Intel's third-quarter earnings report could bring timing updates.

What “Fact-checked” means

Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.

What we checked
Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
How
A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
Who
The checks are made with our newsroom’s technology tools, as steps kept separate from the writing, under rules set by our editor, . A story the checks still flag is not published automatically; it is held for the editor, who decides whether it is fixed, published or dropped.
Archive story
It was written after the event. Its dates were checked against the date of the event, including the dates of any later developments it reports.
If something is wrong
“Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error

How we fact-check →

Companies in this story

IntelPushkar RanadeHot ChipsData center chipsAgentic AI

Earlier coverage of Intel

All Intel coverage →

Written by

Editor · Technology & Business Writer

Hussein is a writer and business technology enthusiast focused on the intersection of technology, entrepreneurship, finance, artificial intelligence, and digital innovation.

CoversAICybersecurityBig TechSaaSStartupsFintech

How this story was made. Researched and written using our newsroom’s technology tools and fact-checked before publication.

Published by Tech CEO Daily, an independent publication. Masthead · Editorial standards

Free newsletters

The technology briefing for people running businesses.

Daily, weekly, bi-weekly or monthly. You choose.

How often

The Daily Brief · Weekdays, 6 a.m. ET

Free forever. One click to unsubscribe. We never sell your email.