TAIPEI – Intel provided more details about its upcoming AI data center graphic processing unit (GPU), codenamed Crescent Island.
The chip will be based on Intel's Xe3P GPU architecture, the company said in a Computex announcement attended by SDxCentral & DCD.
Crescent Island will be a PCI express (PCIe) add-in card with a 350 watt thermal design power (TDP) supporting air cooling. Unlike the latest Nvidia and AMD hardware, which use increasingly supply-limited graphics double data rate (GDDR) or high-bandwidth memory (HBM), the Intel chip uses low-power double data rate 5X (LPDDR5X) memory.
LPDDR5X has lower bandwidth than HBM.
The new GPU will support up to 480-gigabytes (GB) of LPDDR5X memory, although the base reference architecture is 160 GB. The accelerator is expected to be cheaper than higher-end HBM hardware, targeting agentic inference workloads.
The chip supports floating-point four (FP4) for high-performance AI inference, up to FP64, for scientific workloads.
Intel has yet to compute performance figures. Customer sampling of the Crescent Island GPU is targeted for the second half of 2026.
Comments