Shader's do NOT introduce input lag in Retroarch. Raw input driver does not reduce latency as hoped.

[mothergoose729]

I was looking through Brunnis thread in the liberito forums. He has done some valuable testing on the existence of input lag on real hardware vs emulators. He takes the time to test multiple configurations of hardware and OS and his results are fascinating. To summarize:

A real SNES running on a CRT telivision has an input delay of about 3.6 frames, or about 60ms. I have read elsewhere that input lag on real consoles generally range from 2 to 3 frames typically, with Yoshi's Island (the test game) being a bit higher than average. In any event, input lag on a real hardware exists. 

https://forums.libretro.com/t/an-input-lag-investigation/4407/458

Input lag in retroarch is between 4-6 frames (this with vysnc enabled) using BSNES accuracy and GPU hard Sync and no frame delay. This means that in the best case scenario an emulator is within 1 frame of the real hardware, with a 66.8ms-100.2ms input lag range, with 5 frames being the average (83.57ms). 

For reference, streetfighter 5 has 6.5 frames of input lag, and previously had 7.9 frames of input lag on the PS4. 

https://www.eventhubs.com/news/2016/sep/24/street-fighter-5-has-average-input-lag-65-now-according-display-lags-initial-draft-test/

So RA looks pretty good. 

Compared to Higan standalone, retroarch run a lot smoother and has as much as 3 frames less input lag. Worst case scenario for Higan is 7.9 frames input lag. 

In a more recent post, Brunnis compared the difference between running emulators with or without shaders. Using the highly intensive shaders like CRT Geome and CRT-Royale, he noted every test run was within margin of error. Using glsl shades does NOT introduce input lag. 

Unfortunately, using the raw input driver seems to have no benefits over xinput or dinput in his testing. Input lag averages 5 frames in every test. 

https://forums.libretro.com/t/an-input-lag-investigation/4407/458?u=mothergoose729

Something else I found interesting...

Human reaction time to a visual stimulus is between 250-190ms on average, with as few as 1 in 1,000 individuals possessing a sub 100ms reaction time.

https://en.wikipedia.org/wiki/Mental_chronometry

I speculate that the relationship between how fast a human can process and react to what it on screen, and to what degree a person can experience and appreciate input lag, is probably a more complicated and nuanced topic. Interesting nonetheless. 

If RA input lag is consistently within 100ms I think that makes it acceptable for most everyone, and not far off from real hardware in some cases. 

 

 

Edited July 17, 2017 by mothergoose729