Intel Compute Runtime Continues Prepping For GPU D Series, OpenCL LEO for Nova Lake P

Released today was the Intel Compute Runtime 26.40.40196.8 as the newest version of their open-source GPU compute stack for both Windows and Linux systems. With Intel Compute Runtime 26.40.40196.8, a big focus remains on getting the Crescent Island AI accelerator support ready, which is now known as the Intel GPU D Series. With this new Compute Runtime, Crescent Island now has shared system unified shared memory (USM) enabled by default. This allows GPU kernels to access ordinary CPU system memory directly with C/C++ pointers. Defer backing is also now enabled for Crescent Island. Standby mode capabilities are being enabled when paired with the Xe kernel driver only for Battlemage and Crescent Island graphics cards. There are also new performance optimizations within this Intel Compute Runtime update like improving multi-queue usage and increasing usable local memory ring buffer capacity. Intel's OpenCL LEO implementation is also now being enabled by default for Nova Lake P. LEO was introduced earlier this year in experimental form for "Level Zero Executing OpenCL". LEO re-implements OpenCL atop the Level Zero APis rather than having to maintain their own dedicated OpenCL to hardware driver implementation. Great seeing Nova Lake P ready to go the OpenCL LEO route and joins other Nova Lake hardware set to use LEO. There are also some other new features enabled with this Compute Runtime update like a front-end copy capability and an experimental event flags extension. The codebase has also seen some refactoring to optimize compilation for Level Zero. The full list of changes and downloads for the Intel Compute Runtime are available from GitHub.