Benchmarks: OpenCL Kernel Latency ~76x Lower for Lunar Lake with Updated Runtime

This week Intel released the Compute Runtime 25.18.33578.6 release for Windows and Linux. This updated open-source GPU compute stack for OpenCL and Level Zero brings the latest work on Ultra Low Latency Scheduling (ULLS) for Xe2 Lunar Lake graphics and other ongoing Xe2 improvements along with further preparations for next-gen Xe3 hardware. This new Intel Compute Runtime release is clocking in around 76x lower OpenCL kernel latency and other nice wins for those with current-generation Intel Lunar Lake hardware.

This week I carried out benchmarks of the Intel Compute Runtime 25.18.33578.6 to last month's Compute Runtime 25.13.33276.16 as the April release. Testing was done from an ASUS Zenbook S 14 with the Intel Core Ultra 7 256V "Lunar Lake" SoC while running the Linux 6.14 kernel.

The only changes between the benchmark runs were moving from the April release of the Compute Runtime (CR) and the associated Intel Graphics Compiler (IGC) to the new releases this week. No other hardware/software changes were made in just wanting to look at the Lunar Lake Xe2 GPU compute performance impact from these recent optimizations. And, wow, there are improvements indeed.