Intel Compute Runtime 26.05 Released: Supports Nova Lake and Crescent Island accelerators, reports more SYCL code to LLVM

IT House, February 11 — Intel today released the latest Compute Runtime version 26.05.37020.3 and Graphics Compiler (IGC) 2.28.4 updates, fully optimizing Level Zero and OpenCL GPU performance.

This update focuses on the next-generation Nova Lake processors and Crescent Island accelerators, while simultaneously advancing the open-source process for SYCL code in the LLVM mainline.

Core updates for Compute Runtime:

• Completed support for Nova Lake-S processors and updated PCIe device IDs for Nova Lake-U
• Provided driver support for the upcoming Crescent Island AI accelerator cards
• Enabled hierarchical storage mechanisms starting with the Xe2 architecture to improve performance
• Adopted a write-back L1 cache policy for the Xe3 architecture
• Added a Linux IPC Unix socket fallback option
• Supported VR temperature sensor monitoring and multi-GPU memory recommendation interfaces
• Implemented cache tree structure functionality and multi-queue user-space APIs
• Improved TBX host function support and VM binding debug data handling
• Preconfigured Level Zero 1.15 API support framework

This version maintains compatibility with hardware platforms from Tiger Lake to Battlemage, Panther Lake, and provides experimental support for Nova Lake and Crescent Island.

Graphics Compiler upgrade highlights:

• Adapted to LLVM API changes and Clang 17 compiler
• Enabled joint wavefront vectorization by default
• Activated SPIR-V instruction set validation
• Optimized compilation support for Crescent Island accelerators