Intel Plans Crescent Island User Trials by Second Half of 2026
Intel is positioning Crescent Island as part of a broader push to offer end-to-end options across PCs, data centers, and the industrial edge.
Topics
News
- Instagram Tightens Teen Controls Amid Safety Scrutiny
- Anthropic May Store User Data in India as Local Claude Usage Surges
- Intel Plans Crescent Island User Trials by Second Half of 2026
- Oracle Picks AMD to Power New AI Supercluster With 50,000 GPUs
- Oracle Rolls Out AI Database With Built-In Intelligence, Post-Quantum Security
- Google Lines Up $15 Billion AI Data Hub with Adani, Airtel in Vizag
Intel Corp. is preparing a new data-center graphics chip, code-named Crescent Island, aimed at AI inference, the stage where a trained model generates answers to user inputs in real time.
Unveiled at the 2025 OCP Global Summit, the part is slated to reach customers for testing in the second half of 2026.
Crescent Island targets standard air-cooled servers and aims to keep performance high while holding power and costs in check.
The card includes 160GB of on-package LPDDR5X memory, so a larger portion of a model can stay close to the processor, reducing trips to slower external memory and helping maintain steady response times.
Intel said it is focusing on high-volume services that produce large numbers of tokens and other inference-heavy uses where predictable latency and bills matter.
“AI is shifting from static training to real-time, everywhere inference, driven by agentic AI,” said Sachin Katti, CTO of Intel. “Scaling these complex workloads requires heterogeneous systems that match the right silicon to the right task, powered by an open software stack.
The chip uses Intel’s Xe3P design and is intended to sit alongside Xeon 6 processors in the same systems, with support for common AI data types.
Intel is positioning Crescent Island as part of a broader push to offer end-to-end options across PCs, data centers, and the industrial edge.
The pitch centers on familiar server footprints and an open approach that lets operators add accelerators without overhauling cooling or workflows.
Intel is lining up the software before the hardware is available. Its toolchain for mixed CPU-GPU deployments is being exercised on Arc Pro B-Series graphics cards so developers can tune serving stacks now and move workloads when Crescent Island samples arrive.
