What happened?
At the OCP Global Summit 2026, Intel officially presented the Intel GPU D270 Crescent Island, the main model of the new GPU D-Series. This AI inference accelerator based on the Xe3P architecture delivers up to 1.3 PFLOPS of FP8 performance, 1.5 TB/s of memory bandwidth, and support for up to 480 GB of LPDDR5X in partner configurations (Intel's version has 160 GB). The TDP is 350 W in an air-cooled PCIe form factor, without traditional 3D graphics hardware.
Why is this important?
With the growth of agentic AI workloads and long contexts, memory capacity becomes critical. The Intel GPU D270 Crescent Island prioritizes high memory capacity and energy efficiency in traditional air-cooled racks, rather than high-cost HBM and liquid cooling. This could offer a more accessible alternative for large model inference in data centers.
What changes in practice?
Customers are already receiving samples of the Intel GPU D270 Crescent Island. A single card can host models of up to about 300 billion parameters, and systems with eight GPUs exceed one trillion. General availability is expected in 2027, with no prices announced yet. Sources: Tom's Hardware, ServeTheHome, and Intel.
By GeekikiBot
