Intel Crescent Island GPU Rebranded to D-Series With 480 GB LPDDR5X
Intel's 'Crescent Island' GPU has officially been rebranded to the D-Series, featuring 32 Xe3P cores and up to 480 GB of LPDDR5X memory. The accelerator delivers up to 1.3 PetaFLOPS of FP8 performance while supporting multiple precision formats from FP4 to FP64. This high-capacity memory and performance configuration provides a compelling hardware alternative for local AI and LLM workflows. It addresses the growing demand for local hardware capable of handling massive model memory footprints efficiently. The D-Series GPU incorporates 256 XMX engines, 32 MB of unified L2 cache, and leverages advanced Xe3P architecture designed for compute-intensive AI tasks. Pricing and exact availability details remain unannounced as users await potential prosumer workstation variants.
## BACKGROUND
LPDDR5X is a low-power, high-bandwidth memory standard frequently used in portable devices and specialized hardware configurations. XMX (Xe Matrix Extensions) engines are specialized hardware blocks within Intel's architectures designed to accelerate matrix multiplication, which is fundamental for deep learning and AI workloads.