Intel renames Crescent Island as the GPU D-Series and shares targets for FP8 compute and memory bandwidth geared toward AI inference in data centers.
Intel rebrands Crescent Island as the GPU D-Series, a line aimed at AI inference in data centers. It targets 1.3 PFLOPS FP8 and 1.5 TB/s memory bandwidth, and is designed for air-cooled servers with up to eight GPUs per host.
The goal is a simpler, scalable path for multi-tenant AI workloads using a common Xe software foundation. Early samples are in progress, with more details to come from Intel and its partners.
Intel has introduced Crescent Island under a new banner: the GPU D-Series, created specifically for AI inference in data centers. The company pegged a theoretical peak of 1.3 PFLOPS in FP8 and a memory bandwidth of 1.5 TB/s for these accelerators, signaling a sharpened focus on inference workloads.
In practice, the GPU D-Series is built to fit in standard air-cooled rack servers, in a PCIe form factor with a roughly 350W per-device envelope. The design supports configurations of up to eight GPUs per server and aims to handle multi-tenant AI services and long-context inference across models that can reach into trillions of parameters when clustered together.
The hardware is based on Intel’s Xe3P architecture, a direction that emphasizes AI inference rather than traditional graphics. The goal is to offer a predictable, scalable path for deploying AI services and robotic workloads using a consistent software stack across Intel’s Xe family.
Software compatibility is a key part of the plan. Intel described a “common Xe software foundation” to unify drivers and tooling across its Xe cores, which could make it easier for customers to deploy and scale AI workloads across many GPUs. While the company has begun customer sampling, it did not provide a broad launch date or pricing yet.
All of this comes as the industry tries to accelerate AI in the cloud and at the edge. The GPU D-Series is meant to complement Intel’s broader AI hardware strategy by giving data centers a dedicated path to efficient, multi-tenant inference that can scale with models that keep getting larger.
Intel is betting that a focused AI-inference line built on Crescent Island can regain momentum in data centers after past missteps. By targeting multi-tenant inference and long-context workloads, Intel hopes to present a simpler route for customers to run large AI models at scale without bespoke setups for each deployment.
The D-Series is aimed at air-cooled rack servers that support up to eight GPUs per host. Use cases include multiple AI services running in parallel and handling lengthy context from large language models, all in a single, scalable platform.
Intel did not share pricing or a firm release date. It said initial samples are in progress and is building out its ecosystem with OEMs, ODMs, and AI-focused partners. Buyers should expect early tests and partner announcements before any official numbers appear.
For regular readers, this is backend news. If the GPU D-Series leads to faster, cheaper AI services in the cloud or smarter AI features in future Intel products, you’ll notice improvements in the AI features you rely on, from cloud assistants to smarter recommendations in apps that use AI behind the scenes.
Look for broader availability updates, more software details, and independent benchmarks as early samples circulate. The coming weeks and months should reveal how the D-Series stacks up against other inference-focused accelerators.
What is Crescent Island now called?
It’s now the GPU D-Series for AI inference.
What performance targets did Intel share?
1.3 PFLOPS FP8 peak and 1.5 TB/s memory bandwidth are the figures disclosed.
You're reading the quick version.
NVIDIA invites developers and gamers to check out its RTX Spark demos at a series of live stops around the world, with a chance to win a signed high-end laptop.
Meta’s Muse AI agent reportedly generates explicit content and prompts for adult conversations, raising safety and privacy questions for everyday users.
Meta rolls out Muse Wonder and Muse Wonder Full, introducing memory across chats and a looping problem-solving approach as it climbs higher in the AI race.
The Daily Brief
Every morning, the news that matters from AI, phones, apps and the people shaping tech. Explained in plain English. Free.
One email a day. No spam, unsubscribe anytime. Privacy policy
