Key information
At OCP 2026, Intel officially rebranded its Crescent Island AI accelerator as the GPU D-Series. The company disclosed that these chips achieve a theoretical peak performance of 1.3 PFLOPS of FP8 compute and provide 1.5 TB/s of memory bandwidth, marking a significant update in the ongoing disclosures regarding this hardware throughout 2026.
The GPU D-Series utilizes Intel's Xe3P architecture, which is optimized for massively parallel computing rather than traditional graphics. Each unit is designed for an air-cooled PCIe form factor with a 350W TDP and supports up to 480GB of LPDDR5X memory, positioning the hardware for deployment in standard air-cooled rack servers.
Intel intends for these accelerators to handle large language models and latency-sensitive tasks, including multi-tenant AI services and robotics. When scaled to eight GPUs, the system is capable of powering inference for models exceeding one trillion parameters. To support this, Intel is leveraging its common Xe software foundation used across its existing iGPU and discrete GPU product lines.
While customer sampling is currently underway, Intel has not announced a specific date for a broad market launch. The company is collaborating with a wide ecosystem of OEMs, ODMs, and AI-native firms to integrate the technology. The success of the D-Series remains to be seen as Intel seeks to establish a foothold in the competitive AI accelerator market.
