Intel Unveils GPU D-Series AI Accelerator at OCP 2026
Intel introduced the GPU D-Series AI accelerator, an air-cooled chip designed for high-scale AI inference and large language models.
Intel officially named its Crescent Island AI accelerator the Intel GPU D-Series during the OCP 2026 conference. Built on the Xe3P architecture, the chip targets AI inference workloads, providing 1.3 peak PFLOPS of FP8 compute and 1.5 TB/s of memory bandwidth.
The GPU D-Series distinguishes itself from high-power liquid-cooled competitors by utilizing an air-cooled PCIe form factor with a 350W TDP and up to 480GB of LPDDR5X memory. Intel designed the accelerators for deployment in traditional air-cooled rack servers, where they can scale up to eight GPUs to support inference for models exceeding one trillion parameters.
Intel is currently sampling the products to a broad ecosystem of original equipment manufacturers, original design manufacturers, and AI-native companies. The hardware leverages a common Xe software foundation to support latency-sensitive workloads and large language models.
