I've semi-seriously been considering starting my own OS project to circumvent some of the IO latency issues with VR and AR applications. The big metric is moti

I've semi-seriously been considering starting my own OS project to circumvent some of the IO latency issues with VR and AR applications. The big metric is motion-to-photon latency, which is a measure (in milliseconds) of how quickly a user's head motion is reflected on the display. Anything greater than ~10ms is considered insufficient, as you've missed a whole frame of rendering for VR headsets that run at 90hz refresh rate. Current XR applications need to wait for the USB or Bluetooth stack to return the head pose before modifying the view, then wait for the graphics stack to reflect that change. There are many ring transitions along the way. This leads to an absolute minimum motion-to-photon latency that can't be broken just because of OS overhead. This eats into your frame budget, as every ms of context-switching out of the IO drivers leaves you fewer ms to render a frame in time to get it back into the kernel for context-switching back to the graphics driver.