Intel Unveils Crescent Island AI GPU With Up to 480GB LPDDR5X at Computex 2026
Updated
Updated · Tom's Hardware · Jun 1
Intel Unveils Crescent Island AI GPU With Up to 480GB LPDDR5X at Computex 2026
10 articles · Updated · Tom's Hardware · Jun 1
Intel said at Computex 2026 that Crescent Island, its next data-center GPU, will ship as a 350W PCIe card with a reference design carrying 160GB of LPDDR5X and partner configurations scaling to 480GB.
Xe3P is aimed at agentic AI inference, and Intel is betting that high-capacity LPDDR5X can keep more model data on-card while avoiding scarce HBM and advanced packaging bottlenecks.
A fully loaded card could deliver about 684 GB/s of memory bandwidth; an eight-GPU server would reach roughly 3.8 TB of local GPU memory for large models or many smaller agents.
Intel did not disclose throughput or manufacturing details, leaving performance comparisons unresolved as it leans on its oneAPI software stack, which trails Nvidia CUDA and AMD ROCm in adoption.
Crescent Island is slated for the second half of 2026, positioning Intel to target on-premise AI inference systems that can fit into conventional 4U or 5U air-cooled servers.
With Intel pushing CPUs and GPUs for AI, which path will truly challenge Nvidia's dominance?
Is Crescent Island’s unique memory a clever cost-saver or a critical performance bottleneck for AI?
Intel’s Crescent Island AI GPU: 160GB–480GB LPDDR5X Memory Targets Scalable, Affordable AI Inference
Overview
At Computex 2026, Intel unveiled its new AI GPU, Crescent Island, marking a major step in its Data Center Group strategy. This launch reflects Intel’s response to the growing demands of modern workloads, especially those driven by the rapid rise of artificial intelligence. Intel recognizes a fundamental shift in data center requirements, moving away from x86 CPU-only systems to embrace GPU-accelerated training and inference. The company now focuses on maximizing performance-per-watt, per-core efficiency, and memory bandwidth, aiming to meet the evolving needs of AI-powered data centers and support the next generation of AI applications.