AI-rewritten: This is a summary of an article from Tom’s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

At OCP 2026 today, Intel officially named its Crescent Island project the Intel GPU D-Series. The company shared theoretical peak performance targets of 1.3 PFLOPS for FP8 and a memory bandwidth of 1.5 TB/s. These figures represent the first time Intel has disclosed specific performance metrics for the chip, adding to the gradual information released about Crescent Island throughout 2026.

The GPU D-Series is built on Intel’s Xe3P graphics architecture, which functions as a massively parallel computing system rather than a traditional GPU due to its lack of standard graphics functional units. The new products will use an air-cooled PCIe form factor with a thermal design power (TDP) of 350W and up to 480GB of LPDDR5X memory. This contrasts sharply with Nvidia’s Rubin chips, which require liquid cooling and operate at kilowatt-level power consumption.

Intel states that the GPU D-Series targets traditional air-cooled rack servers capable of housing up to eight GPUs. In this configuration, the chips could support inference on models exceeding one trillion parameters. The company highlights suitability for large language models, long-context inference, and concurrent services, aiming to enable multi-tenant AI applications in robotics and latency-sensitive workloads.

Intel is developing a "common Xe software foundation" to manage these distributed systems, working with OEMs, ODMs, neoclouds, and independent software vendors. While customer sampling is currently underway, no broad launch details have been provided. The product aims to address previous strategic challenges Intel faced with its Data Center GPU Max family and failed silicon acquisitions.