Intel names its inference-focused data center GPU 'GPU D-Series'; D270 has 480GB LPDDR5X, 350W, air-cooled (reported)

Verified (fact)

At the 2026 OCP Global Summit on October 12 (local time), Intel revealed that its Xe3P-based inference GPU 'Crescent Island' will officially be called 'Intel GPU D-Series.' Specs of the top model, D270: - Memory: up to 480GB LPDDR5X - Power and cooling: 350W, air-cooled - Memory bandwidth: up to 1.5TB/s - FP8 compute: up to 1.3 PFLOPS The release date is not given in the reports.

Needs fact-check (claim)

Intel said a single card can run models of up to 300 billion parameters, and eight cards can run 1-trillion-parameter models. It also said open-source AI models and PyTorch will be supported out of the box from launch day.

Why it matters (interpretation)

By choosing high-capacity LPDDR5X and air cooling instead of HBM, Intel targets the inference market, where capacity, power and cost matter more than bandwidth. The configuration could serve as an alternative amid the HBM supply crunch.

Sources

• Press 인텔, 추론 특화 GPU 정식 명칭 공개... "GPU D시리즈" ZDNet Korea