Intel's next-gen 'Crescent Island' data center GPUs: new Xe3P architecture, up to 160GB LPDDR5X

Intel confirms new Xe3P GPU architecture isn't just for integrated graphics: Crescent Island inference-optimized data center GPUs use Xe3P GPU.

Intel's next-gen 'Crescent Island' data center GPUs: new Xe3P architecture, up to 160GB LPDDR5X
Comment IconFacebook IconX IconReddit Icon
Gaming Editor
Published
1 minute & 15 seconds read time
TL;DR: Intel's new Crescent Island GPU, launching in 2026, features the advanced Xe3P architecture and up to 160GB LPDDR5X memory, optimized for power-efficient AI inference workloads in air-cooled data centers. It offers cost-effective performance-per-watt, targeting enterprise servers and token-based AI services.

Intel has announced its new Crescent Island GPU for data centers, featuring the next-gen Xe3P architecture, and up to 160GB of LPDDR5X memory for AI workloads.

Intel's next-gen 'Crescent Island' data center GPUs: new Xe3P architecture, up to 160GB LPDDR5X 71
2

The new Intel Crescent Island GPU is based on the new Xe3P architecture, which is the same GPU architecture that is going into Intel's next-gen Panther Lake laptop processors in 2026. Xe3P is another upgrade on top of the Xe3 GPU architecture, first inside of Crescent Island, and then on the consumer side with the new Arc C-series GPUs with Nova Lake next year.

Intel explains: "The new data center GPU code-named Crescent Island is being designed to be power and cost-optimized for air-cooled enterprise servers and to incorporate large amounts of memory capacity and bandwidth, optimized for inference workflows".

Intel's new Crescent Island GPU will be power- and cost-optimized, aiming for air-cooled data centers and ready for AI inference workloads. Intel says that its new Xe3P graphics architecture inside of Crescent Island is optimized for performance-per-watt, with the card sporting 160GB of LPDDR5X memory, a change from the higher-end HBM memory used by AI GPU competitors NVIDIA and AMD.

NVIDIA and AMD are currently using both HBM3 and HBM3E memory on their AI chips, with both companies shifting to next-gen HBM4 memory for Rubin, and Instinct MI400, respectively. LPDDR5X is a cheaper standard for Intel to use, but we can't wait to see what performance is like when Crescent Island launches in 2026.

Intel "Crescent Island" data center GPU key features include:

  • Xe3P microarchitecture with optimized performance-per-watt
  • 160GB of LPDDR5X memory
  • Support for a broad range of data types, ideal for "tokens-as-a-service" providers and inference use cases