Open Source Driver Renders Windows 11 Desktop Using NPU Instead of GPU

A developer group known as labscommunity has released a prototype open-source driver that enables a Neural Processing Unit (NPU) to render the Windows 11 desktop. The project, released under an Apache 2.0 license on GitHub, aims to offload desktop composition from the system's integrated or discrete GPU to the NPU. Testing conducted on Intel Panther Lake systems confirmed the NPU can successfully composite the interface, including applications such as Edge and Notepad, without the compositor or applications requiring modification.

While the technical proof-of-concept was successful, current performance remains significantly below usable levels. The driver currently operates at a refresh rate of only 2 to 3 Hz, with some tests reporting speeds as low as 0.6 FPS. The developers noted that the project is strictly "a research prototype, not a product" at this stage. The group has set an ambitious performance target of 30 Hz for future revisions of the software.

Despite the low frame rates, the developers reported that the driver can reduce 3D GPU usage from 3.4% to 0.6%, potentially freeing up the GPU for other tasks such as inference. To prevent system crashes, the driver includes fail-safes that allow it to automatically fall back to the standard GPU if a software fault occurs or if a user requests it.

According to the project's readme file, the development was motivated by the fact that many modern laptops ship with NPUs that remain idle while the GPU handles desktop composition and browser tasks. The project seeks to test whether the NPU can take over these responsibilities end-to-end. The current testing utilized Intel Panther Lake processors featuring NPU 5 architecture with performance spanning 45-50 TOPS.

Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.