Intel Launches Crescent Island An AI Inference GPU Dedicated to Data Centers
Intel has announced a new data center GPU Crescent Island, which would be focused on workloads associated with AI inference. The new product utilizes the yet-to-be-launched Xe3P graphics architecture and intends to be an economical option in terms of power and cost for corporate servers.
Xe3P Architecture and Made Features
Powering the Crescent Island GPU is the new Xe3P architecture-an uplifted version of the Xe3 graphics architecture designed to support future client products like the Arc C-Series. The Xe3P will be set up for high scalability ranging from integrated client graphics to dedicated data center AI GPUs.
Essential features of the Crescent Island GPU include:
- Architecture Xe3P microarchitecture designed with high-performance tight-watt optimized for AI inference work.
- High Memory Capacity Has 160GB LPDDR5X memory, offering a different approach to the HBM memory used by competitors.
- Broad Data Type Support Ideal for "tokens-as-a-service" providers and other types of inference use cases.
- Air-Cooled Design Optimized for saving power and cost in standard air-cooled enterprise server environments.
Strategy and Market Position
Here, the message is clear that it plans to go for cost economics with LPDDR5X memory instead of taking on the more expensive HBM memory that is at limited supply. Possible positioning for Intel will thus be competitive in both cost and availability.
Currently, the company is developing and testing its open and unified software stack on current Arc Pro B-Series GPUs, which will allow early optimization of software and iteration into Crescent Island when it's ready.
When Will It Be Available
Expected customer sampling of the Crescent Island GPU would commence in the latter half of 2026. Further architecture and product specifications are set to be done closer to its sampling.
