Intel this week released their last planned feature update to their open-source Compute Runtime for 2025. The Intel Compute Runtime 25.48.36300.8 delivers the latest OpenCL and Level Zero performance optimizations, Xe3 workarounds, and other fixes for those on Intel integrated and discrete graphics hardware.
Intel Compute Runtime 25.48.36300.8 incorporates all of the features and fixes worked on by Intel software engineers since early November. Some of the Intel Compute Runtime highlights for this update include:
– Exposing Level Zero API version 1.14.
– Updating the product configuration for the AOT Arrow Lake H B0 target.
– Xe3 fixes like enabling a state cache invalidation workaround by default.
– Optimizing the img2buf and bug2img built-ins.
– A new SLM access protocol feature.
– ISA allocation pooling in OpenCL is now implemented.
– Dropping the GNU compiler version checks for when to not use Link Time Optimizations (LTO) for the compiler when building the Compute Runtime. The compiler LTO support will now be assumed to use without checking for older versions of GCC.
– Various performance optimizations like timestamp caching on OpenCL and Level Zero. Plus other performance work like Level Zero host Unified Shared Memory (USM) growing pools.
– Various bug fixes.
The Intel Compute Runtime continues supporting back to Tiger Lake integrated graphics with OpenCL 3.0 and Level Zero 1.14 up through the latest Lunar Lake / Arrow Lake and Battlemage graphics. Plus pre-release support for Panther Lake SoCs too.
Downloads and more details on the Intel Compute Runtime for Linux and Windows systems via GitHub.
