Intel GPU D-series

The Intel GPU D-series (formerly code-named "Crescent Island") is a line of data center accelerators purpose-built for efficient, large-scale AI inference. Announced in October 2026, the series is designed to handle memory-intensive workloads like large language models (LLMs) and long-context tasks without the need for specialized liquid-cooling infrastructure.

Key Specifications (D270 Flagship Model)

The flagship D270 is the first announced card in the series. Its specifications prioritize memory capacity and power efficiency for enterprise environments:

• Architecture: Based on the Xe3P architecture, which reallocates silicon area from graphics and ray-tracing pipelines to focus exclusively on matrix computation and AI processing.
• Memory: Supports up to 480GB of LPDDR5X memory (via partner ODM configurations; Intel’s reference card features 160GB).
• Performance: Rated at up to 1.3 petaflops of theoretical FP8 matrix performance.
• Bandwidth: Provides 1.5 TB/s of memory bandwidth.
• Power & Cooling: Operates within a 350-watt power envelope and utilizes a standard air-cooled PCIe 5.0 form factor.
• Compute Cores: Features 32 Xe3P cores and 256 XMX (Xe Matrix Extensions) engines.

Core Purpose and Positioning

• Target Workloads: The D-series is optimized for "agentic" AI, long-context inference, and concurrent AI services.
• Model Scaling: Intel claims a single card can support models with up to 300 billion parameters, while an eight-GPU system can handle models exceeding one trillion parameters.
• Deployment: By using an air-cooled, standard PCIe design, the D-series is intended to be easily integrated into existing data center server infrastructure, avoiding the high cost and complexity of liquid-cooled racks required by some high-end competitors.

As of October 2026, customer sampling is underway. Intel has not yet disclosed pricing, broad availability dates, or independent performance benchmarks.