Intel introduced its new data center GPU, codenamed “Crescent Island,” at the 2025 OCP Global Summit. Optimized for AI inference workloads, the GPU features the Xe3P microarchitecture and integrates 128GB of LPDDR5X memory, targeting high efficiency and cost-effectiveness in air-cooled enterprise servers.
Intel’s CTO highlighted the industry shift from training to inference, emphasizing the need for open software stacks and heterogeneous systems. The GPU supports multiple data types and is designed for scenarios like tokens-as-a-service, with samples expected in the second half of 2026.
The company is currently developing its software stack using the Arc Pro B-series GPU, aiming to accelerate AI inference deployment across diverse applications.
ICgoodFind : Intel’s large-memory GPU underscores the critical role of hardware in advancing AI inference, heating up competition in the specialized chip segment.
