At the Open Compute Project Global Summit, Intel unveiled technical details of its GPU D-series AI accelerator, codenamed Crescent Island. [1, 2, 3] Built on Intel’s latest Xe3P architecture, the chip is designed for AI inference workloads, including running large language models. [4, 7, 11]
The accelerator uses a standard air-cooled PCIe card format, allowing businesses to upgrade their AI capabilities without investing in specialized liquid cooling or new rack systems. [1, 3, 5, 7] It delivers up to 1.3 petaflops of performance. [2, 5] In a multi-GPU setup, it can handle models with more than one trillion parameters. [2, 5] Customer sampling is underway. [1, 2]