Intel has officially unveiled its new D270 GPU, based on the Xe3P architecture, specifically designed for AI inference tasks rather than gaming. The company presented the hardware at the Open Compute Project 2026 summit.
The D270 processor delivers up to 1.3 PFLOPS of performance in FP8 format and offers a memory bandwidth of 1.5 TB/s. It features 32 Xe3P cores, 256 XMX blocks, and 32 MB of unified L2 cache.
To maximize AI performance, Intel removed traditional 3D graphics and ray tracing hardware blocks from the chip. The reference design uses a PCIe 5.0 interface with 160 GB of LPDDR5X memory, though partners can scale this up to 480 GB.
The accelerator has a TDP of 350 W and utilizes passive air cooling compatible with standard server infrastructure. Intel states that a single unit supports AI models with up to 300 billion parameters, while an eight-GPU system can handle models exceeding one trillion parameters.
Sample shipments to customers have already commenced. Intel has not yet disclosed pricing details or the official timeline for a broad market release.
