Intel names its Crescent Island AI chip: GPU D-Series, with up to 480 GB of memory

Intel used the OCP Global Summit to give its data-centre inference GPU, codenamed Crescent Island, a product name: Intel GPU D-Series. The first model is the GPU D270.

Specs at a glance

• Architecture: Xe3P, with 32 Xe3P cores and 256 XMX matrix engines.
• Memory: LPDDR5X, up to 480 GB on partner cards. Intel's own card has 160 GB.
• Bandwidth: 1.5 TB/s.
• Compute: 1.3 PFLOPS FP8 peak.
• Power and form factor: 350 W, air-cooled, PCIe 5.0 x16.

Using LPDDR5X instead of expensive HBM is a deliberate choice. It gives up some bandwidth in exchange for large capacity at a lower cost, which suits serving big models to many users.

Racks too

Intel also showed new Xeon rack designs based on the OCP ORv3 standard, part of an "open infrastructure" push for agentic AI.

Availability

Customer sampling is under way. Intel has not given pricing or a general-availability date.