Diamonds, Crescents and Wildcats: Intel shows off its hardware for the next generation of agentic AI workloads | Tech Radar
Overview
News, deals, reviews, guides and more on the newest computing gadgets
Start exploring exclusive deals, expert advice and more
Details
Unlock and manage exclusive Techradar member rewards.
Unlock instant access to exclusive member features.
Get full access to premium articles, exclusive features and a growing list of member rewards.
Diamonds, Crescents and Wildcats: Intel shows off its hardware for the next generation of agentic AI workloads
When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.
Intel shows three agentic AI architectures at Hot Chips 2026: the 256-core Diamond Rapids Xeon, the Crescent Island inference GPU, and the already-shipping Wildcat Lake client So C
Diamond Rapids matches AMD's EPYC Venice on cores and bandwidth but arrives in 2027 with half the threads, while Wildcat Lake's 17 TOPS NPU falls short of Copilot+'s 40+ TOPS certification
Intel's Crescent Island will support up to 480 GB via third-party partner cards, but the Intel-branded variant will be limited to 160 GB, even as it currently does not publish the memory bandwidth that determines its token economics.
Intel used its slot at Hot Chips 2026 to lay out three architectures it says will carry agentic AI from the rack to the laptop.
The already-shipping Wildcat Lake Core Series 3 So C is for mainstream client and edge machines, Crescent Island is an inference-optimized data center GPU, and Diamond Rapids is its next-generation enterprise-class data center CPU that can be configured with up to 256 cores.
Pushkar Ranade, Intel's chief technology officer, said in a statement accompanying the announcement that agentic AI is changing how the company designs computing from the transistor upward.
Plenty of compute promises, sans some GPU numbers
Intel's Diamond Rapids, which it plans to brand as Xeon 7, is where the chipmaker provides the most information, unsurprisingly, in an industry that plans purchases and upgrades years in advance.
Its flagship offering can encompass as many as 256 cores, backed by 1.28G of fast last-level cache, 16 memory channels, 128 PCIe 6 lanes, and CXL 3.0. Serve The Home reported that the maximum configuration uses 16-core chiplets, built on Intel's 18A-P process node.
‘Those two jobs need different physics’: Rebellions CEO says training and inference need different chips
Why Arm is set to help power the next generation of AI workloads — and beyond
Intel says it 'welcomes the competition' from Nvidia's new RTX Spark CPU
Rival chipmaker AMD also offers a 256-core trim: the EPYC 9996, built on TSMC's 2nm node, keeps hyperthreading in play, effectively doubling its thread count versus Intel's flagship.
Its Crescent Island GPU is arguably the most interesting design of the three, but the least detailed: a 350W air-cooled PCIe card built on its Xe 3P GPU architecture, with 32 cores feeding 256 third-generation XMX engines and 32 MB of unified L2 cache. It surprisingly uses LPDDR5X rather than HBM, but the choice seems rationalized to offer capacity at a power and cost profile that lets Intel compete meaningfully with Nvidia and AMD.
Interestingly, the 480GB figure Intel touts in its press release isn't what it will ship; Intel-branded cards cap at 160GB, and 480GB is a ceiling available to ODM partners building their own configurations. Intel also remains mum on memory bandwidth specifics and told Chips and Cheese it won't disclose the figure; this is crucial for many AI applications, as memory bandwidth determines decode throughput, and decode throughput essentially determines token economics.
Of the three architectures that Intel talked about, Wildcat Lake is the only one you can pick up off a shelf (as a mini PC or laptop) today; it launched as Intel's Core Series 3. Intel stated that it replaced the Foveros packaging used in Panther Lake with an organic multi-chip package, eliminating the base die and, with it, assembly cost and yield loss. Compute die area fell by 38 percent, and I/O tile area by 15 percent.
The 17 TOPS NPU available to such CPUs sits well below the 40 TOPS Microsoft requires for Copilot+ certification. The 40 TOPS Intel cites is a platform total across CPU, GPU, and NPU, which arguably can be met elsewhere, but it's a sticking point for the chipmaker's AI push, which has otherwise been concentrated on capturing enterprise and hyperscale customers.
Follow Tech Radar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
Having built hundreds of gaming PCs and being an avid gamer in his spare time, Rahim tends to have stronger opinions about hardware than most. This is particularly on display when he gets his way with powerful, but minimalistic RGB builds even as Small Form Factor (SFF) PCs come a close second.
You must confirm your public display name before commenting
Soundcore Nebula P1 review: a travel-friendly projector that stands out from the crowd
The US Navy wants its next drone-mounted torpedo to be able to glide silently through the air
"The best way to predict the future is to invent it" — quote of the day by physicist Dennis Gabor
Ukraine has built its own jet-powered drone interceptor — and it'll easily fit in the back of a pickup truck
Russia says it will start selling its 'suicide drone' to foreign nations — but won't name them
Tech Radar is part of Future US Inc, an international media group and leading digital publisher. Visit our corporate site.
© Future US, Inc. Full 7th Floor, 130 West 42nd Street, New York, NY 10036.
Key Takeaways
- News, deals, reviews, guides and more on the newest computing gadgets
- Start exploring exclusive deals, expert advice and more
- Unlock and manage exclusive Techradar member rewards
- Unlock instant access to exclusive member features
- Get full access to premium articles, exclusive features and a growing list of member rewards
