Intel Unveils Crescent Island AI Data Center GPU: Up to 480GB LPDDR5X, Air-Cooled for Inference
June 1, 2026
Intel Unveils Crescent Island AI Data Center GPU: Up to 480GB LPDDR5X, Air-Cooled for Inference
Intel has provided new details on its upcoming AI data center GPU, codenamed Crescent Island, positioning it as a cost-effective alternative for inference workloads in an increasingly memory-constrained market. The announcement, made during a Computex presentation attended by DCD, highlights Intel's strategy to differentiate through memory architecture and thermal design.
The chip is built on Intel's Xe3P GPU architecture and is designed as a PCI Express add-in card with a 350W thermal design power (TDP), supporting air cooling. This stands in contrast to the latest offerings from Nvidia and AMD, which rely on increasingly supply-limited GDDR or HBM memory. Instead, Crescent Island uses LPDDR5X memory, which, while offering lower bandwidth than HBM, enables a significantly larger memory pool at a lower cost.
The GPU will support up to 480GB of LPDDR5X memory, though the base reference architecture starts at 160GB. Intel expects the accelerator to be cheaper than higher-end HBM-based hardware, targeting agentic inference workloads—a growing segment that demands high memory capacity but can tolerate lower bandwidth. The chip also supports a wide range of precision formats, from FP4 for high-performance AI inference up to FP64 for scientific computing workloads.
Intel has yet to disclose specific performance figures. Customer sampling of the Crescent Island GPU is scheduled for the second half of 2026, suggesting a potential commercial launch in late 2026 or early 2027. The move underscores Intel's push to capture a share of the AI inference market, where memory capacity and cost efficiency are becoming critical differentiators.
Source: datacenterdynamics
