Intel Renames Crescent Island AI Accelerator to GPU D-Series, Details Performance

Official Naming and Performance Specifications

Intel has officially named its AI accelerator, previously known as Crescent Island, the Intel GPU D-Series. The company also provided specific performance targets for the chip, including a theoretical peak performance of 1.3 PFLOPS for FP8 operations and a memory bandwidth of 1.5 TB/s. These details expand on prior disclosures about the accelerator.

Architecture and Form Factor

The GPU D-Series is built on Intel's Xe3P architecture, which is described as a massively parallel computing architecture. These products will be available in an air-cooled PCIe form factor, featuring a Thermal Design Power (TDP) of 350W and up to 480GB of LPDDR5X memory. The design indicates a focus on deployment within existing data center infrastructure.

Target Market and Use Cases

Unlike high-power, liquid-cooled solutions, the Intel GPU D-Series is designed for traditional air-cooled rack servers, supporting configurations with up to eight GPUs. Intel states that these chips can power inference for models exceeding one trillion parameters. Specific applications include large language models, long-context inference, concurrent inference services, multi-tenant AI services, robotics, and other latency-sensitive workloads.

Software and Ecosystem Support

Intel emphasizes its common Xe software foundation for AI across its integrated and discrete GPU products, which will support distributed inference deployments for the GPU D-Series. The company is collaborating with a broad ecosystem of OEMs, ODMs, neoclouds, independent software vendors, appliance providers, and AI-native companies to facilitate adoption. Customer sampling of the GPU D-Series is currently underway.