Intel's Crescent Island AI accelerator officially powers new GPU D-Series products
Intel's Crescent Island AI inference accelerator will come to market as the GPU D-Series family, promising 1.3 PFLOPS of FP8 compute alongside up to 480 GB of LPDDR5X delivering 1.5 TB/s of memory bandwidth.
Intel has officially unveiled its new AI accelerator, the Intel GPU D-Series, at OCP 2026. The chip is built on Intel's Xe3P graphics architecture, specifically designed for AI inference workloads. With a theoretical peak performance of 1.3 PFLOPS of FP8 and a final memory bandwidth of 1.5 TB/s, the D-Series aims to deliver powerful inference capabilities without the need for liquid cooling or high TGP, unlike Nvidia's Rubin GPUs.
The GPU D-Series will be available in an air-cooled PCIe form factor, supporting up to 480GB of LPDDR5X memory and suitable for deployment in traditional rack servers with up to eight GPUs. Intel emphasizes the chip's potential for powering large language models, long-context inference, and concurrent inference services. The company is working on a robust software stack, the "common Xe software foundation," to enable multi-tenant AI services, robotics, and other latency-sensitive workloads.
Intel is collaborating with various partners, including OEMs, ODMs, neoclouds, independent software vendors, appliance providers, and AI-native companies, to bring the Xe accelerators to the data center. Sampling of GPU D-Series products is currently underway, with a potential broad launch yet to be determined. As AI models continue to advance, the Intel GPU D-Series could provide a crucial AI accelerator foothold for Intel, addressing the increasing demand for inference compute cycles and catering to the requirements of increasingly intelligent and agentic AI workflows.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
