Here's ezQuake doing 960fps at 144Hz generating approximately (960/144) = about 6-7 tearlines per refresh cycle. Lower the refresh rate to 60 Hz, and this turns into 16-17 tearlines per refresh cycle with even tinier offsets.
That unfortunately means roughly 85% of rendering is wasted in order to achieve guaranteed subrefresh latency, since full frames are rendered, only to stream a thin frameslice out of the GPU on the spot, for an extremely brief moment -- but what you gain is true subrefresh button-to-photons latency on some gaming monitors that does realtime cable-to-panel synchronous scanout.
If you properly enable A+B+C+D in my post -- and you use a monitor capable of subrefresh latency -- then it is already confirmed that the topmost frameslices photons are already hitting your eyes, long before even the bottommost frameslices are even begun to be rendered!
Even the bottommost frameslices photons can be beginning to hit your eyes when the topmost frameslice of the next refresh cycle is begin to be rendered. And even the wraparound frameslice (one frameslice that wraps around from bottom edge back to the top edge).
The refresh cycle thus is a metaphorical rolling ring buffer of latency, with the frameslices streamed at whatever point the raster is currently at. As long as you're VSYNC OFF, and as long as the game is blasting frames faster than Hz, you're
already getting subrefresh latency button-to-GPU-output.
Effectively, you're getting accidental beamraced lag benefits simply from the sheer overkill framerate and sheer render wastage, just to provide an "low-lag-at-raster" effect since a fully rendered frame is available, so there's low-lag scanlines already available to deliver at whatever raster point. Now, going even excessively further we go 6000fps at 60Hz, we're wasting 99% of each frame, but 1% of each 6000 frames (except for those frameslices fitting in VBI) is made human visible -- at nearly 100 frame slices per refresh cycle. But we only need 300fps at 144Hz to begin to see accidental beamraced lag savings benefits that may not be factored in.
Unfortunately, not many researchers understand this, factoring all of these factors in. Researchers are often specialized experts for their respective areas -- the Olympics athlete -- and when they try to lag-measure screens, they don't understand the subrefresh nuances, lag gradients, and the interactions between GPU and monitor. But a game running VSYNC OFF at framerates far in excess of refresh rates -- is apparently giving accidental beamraced lag-reducing benefits by the sheer defacto effective streaming of frameslices in real time.
Assuming A=true, B=true, C=true, D=true, then game has always (historically) been found capable of giving sub-refresh latency from button-to-GPU-output, no matter what the game is. Simply, A+B+C+D simply needs to be true, in order to be confident about being able to achieve subrefresh latencies.
(A) VSYNC OFF
(B) fps > Hz by significant amount
(C) Mouse raw input, atomic events, choose a known low-lag mouse (to avoid laggy mice)
(D) Software rendering (not hardware sprites, e.g. mouse cursor)
Obviously, you'll have to choose a display that has subrefresh latency ability (using VSYNC OFF lag measurement metric, aka first-anywhere metric, rather than Leo Bodnar scanout-latency-affected measurement metric). But choose an "esports mouse" and an "esports display", they will usually (>50%-95% of the time) have low enough latencies not to be a human-reaction-time error margin. Even randomly selected 240Hz monitors and randomly selected 1000Hz mice, now tends to affect lag chain results only by less than a 10ms amplitude, which is impressively tight. At least mouse lag is equipment-measurable, and display lag is equipment-measurable, so these can be controllable.
That said, I think we need more esports research that factors.
- scanout and understanding how VSYNC OFF becomes accidental beamracing
- first-anywhere vs first-singlepoint
- stimuli differences (visual flash vs audio gunshot) and multi-stimuli that often leapfrog each other (flash, movement, audio, direct-gazes, peripherals, etc) in ways that low-end scientific equipment is not always measuring.
- the hyper andrenaline factor (esports gamers are often really pumped like a skydiver at their peak)
- researching unexpected predictive factors (identifying unidentified earlier stimuli that may be occuring before the stimuli, such as onscreen ammo number decrementing before gunshot flash, etc)
Etc.