Intel Enabling Ultra Low Latency Scheduling "ULLS" For Lunar Lake GPU Compute
While last week Intel released an update Compute Runtime for GPU compute with the OpenCL and Level Zero APIs on Windows and Linux, today they released a new preview version for readying a shiny new feature: Ultra Low Latency Scheduling "ULLS" for Lunar Lake Xe2 graphics.
Intel engineers have been working on Ultra Low Latency Scheduling "ULLS" as a feature for a while to allow direct submission of work to the GPU for bypassing some of the driver overhead and helping with lower latency for compute kernels. ULLS is also referred to just as Direct Submission and with today's preview release is enabled by default for Lunar Lake graphics on Linux. Other vendors have also been working on similar functionality such as the AMD Linux driver with user mode queues.
The Compute Runtime also supports the notion of ULLS Light as Ultra Low Latency Scheduling without the VM_BIND kernel interface. ULLS Light in turn has been supported with parts like Meteor Lake H and Arrow Lake H.
In turn following today's preview release, Lunar Lake Direct Submission / ULLS is expected to become part of the official Compute Runtime release due out in May with Compute Runtime 25.18.
Intel Lunar Lake on Linux continues to enjoy support improvements, can deliver quite nice single-threaded performance and great energy efficiency. Lunar Lake SoCs have wound up in interesting and quality laptop designs such as the recently tested Lenovo ThinkPad X1 Carbon Gen 13 Aura under Linux. Plus it makes a good choice for those wanting official GPU compute support with the Intel Compute Runtime compared to unofficial support for mobile parts with the likes of AMD ROCm.
Those wanting to try out this new feature enabled by default for helping with better latency and in turn performance for Lunar Lake hardware can find the pre-release via Intel Compute Runtime 25.13.33276.19. I'll be working on some ULLS comparison benchmarks soon.
Intel engineers have been working on Ultra Low Latency Scheduling "ULLS" as a feature for a while to allow direct submission of work to the GPU for bypassing some of the driver overhead and helping with lower latency for compute kernels. ULLS is also referred to just as Direct Submission and with today's preview release is enabled by default for Lunar Lake graphics on Linux. Other vendors have also been working on similar functionality such as the AMD Linux driver with user mode queues.
The Compute Runtime also supports the notion of ULLS Light as Ultra Low Latency Scheduling without the VM_BIND kernel interface. ULLS Light in turn has been supported with parts like Meteor Lake H and Arrow Lake H.
In turn following today's preview release, Lunar Lake Direct Submission / ULLS is expected to become part of the official Compute Runtime release due out in May with Compute Runtime 25.18.
👁 Lenovo ThinkPad X1 Carbon Gen 13 Aura
Intel Lunar Lake on Linux continues to enjoy support improvements, can deliver quite nice single-threaded performance and great energy efficiency. Lunar Lake SoCs have wound up in interesting and quality laptop designs such as the recently tested Lenovo ThinkPad X1 Carbon Gen 13 Aura under Linux. Plus it makes a good choice for those wanting official GPU compute support with the Intel Compute Runtime compared to unofficial support for mobile parts with the likes of AMD ROCm.
Those wanting to try out this new feature enabled by default for helping with better latency and in turn performance for Lunar Lake hardware can find the pre-release via Intel Compute Runtime 25.13.33276.19. I'll be working on some ULLS comparison benchmarks soon.