News AMD Anti-Lag steps out of its comfort zone — tech arrives on Vulkan 1.3.291, bringing Anti-Lag benefits outside of DirectX for the first time

Reflex/Anti-Lag are not 'FPS capping'. What they are doing is talking to the game engine and telling it "the previous frame has not started scanout yet, wait before you submit another frame". This means that the render ideally never exists, as there is no render to buffer - as soon as the previous frame starts canning out, the engine can then take the most recent input sample (mouse position etc), computer the frame, submit the frame to the GPU, and render the frame. The caveat is that for this to work, the engine has to be sure that the end-to-end frame delivery time will always be shorter than the interval between frame deliveries - the basic reciprocal of the monitor refresh rate, e.g. 16.67ms at 60Hz. If it takes longer, you mish the next scanout start and get a 1-frame stutter. This makes reflex/anti-lag great if you can hold minimum FPS (really maximum frametime) well above the monitor refresh rate, but not much use if you are ever dipping below that.
In effect, this clamps latency to the reciprocal of the monitor refresh rate.

For VR, you can do 'late latching', which is where you not only wait for the next start[ of scanout, but by packing frame delivery (knowing pretty well how long a given frame will render) you wait to even start generating that frame until the absolute last possible moment before it would result in a render miss. This means the minimal latency can be BELOW the reciprocal of the refresh rate, e.g. if you could render 240FPS but your monitor can only display at 60Hz, rather than uncapped framerate (4-frame tearing across screen) or VSYNC (60FPS), the game will pause render start until 4.17ms before the finished frame needs to start readout, and only start rendering then. The tradeoff is that in the event of a render miss, the 'stutter' impact is not a BIG spike in latency (going from your low-latency target to double your refresh rate reciprocal - 4.17ms to 33.33ms for our example). VR tackles this by reprojecting every single rendered frame such that head movement always has the minimum possible latency regardless of render latency, even though world objects may be subject to latency in motion. For flat games, this would correspond to having rock solid low mouse movement latency, but character animations still subject to latency judder.