Intel has initiated customer sampling for its new D-Series data center GPUs, previously codenamed Crescent Island. The flagship D270 model is built on the Xe3P architecture and is designed specifically for AI inference and agentic workloads rather than model training.
The D270 features 32 Xe3P cores, 256 XMX engines, and a 32 MB L2 cache. It delivers a peak compute performance of approximately 1.3 PFLOPs at FP8 precision. The chip utilizes an air-cooled design and a PCIe Gen5 form factor for compatibility with standard servers.
A key feature of the D270 is the use of LPDDR5X memory instead of HBM. The reference configuration includes 160 GB of memory, which can be expanded to 480 GB by partners, with a peak bandwidth estimated at 1.5 TB/s.
Intel aims to integrate these GPUs with its Xeon processors to provide comprehensive AI inference systems. OEM partners for the initiative include ASUS, Dell, and HPE. The company is positioning the hardware as a cost-effective alternative to solutions from NVIDIA and AMD.
The broader customer sampling phase is scheduled for the second half of 2026. Intel previously showcased the technology at Computex and Hot Chips 2026 to prepare for the product launch.
