Mouse latency test reveals 2x higher latency than normal, what do I do?

[jorimt]

Also the game was at 300fps. And the samsung camera records 0.4s @960fps and then plays it back at 30fps, I don't know if that had anything to do with it. And sorry the video won't display, you'll have to click the link.

I fixed your video embed; you didn't have the tags wrapped around the url.

As for your video, since you're testing V-SYNC off at such high framerates and you don't have any horizontal movement visible during capture, assuming you're only watching for the muzzle flash (maybe you're not, you didn't say), you may be missing the first on-screen reaction to your input the majority of the time, which would increase your recorded averages.

Those tear slices get small, and the initial update with no sync can occur anywhere on the screen (and be very subtle). See:
https://forums.blurbusters.com/viewtopi ... 510#p52233



Also, why do you think you should be getting ~16ms?

[LLm1]

Also the game was at 300fps. And the samsung camera records 0.4s @960fps and then plays it back at 30fps, I don't know if that had anything to do with it. And sorry the video won't display, you'll have to click the link.

I fixed your video embed; you didn't have the tags wrapped around the url.

As for your video, since you're testing V-SYNC off at such high framerates and you don't have any horizontal movement visible during capture, assuming you're only watching for the muzzle flash (maybe you're not, you didn't say), you may be missing the first on-screen reaction to your input the majority of the time, which would increase your recorded averages.

Those tear slices get small, and the initial update with no sync can occur anywhere on the screen (and be very subtle). See:
https://forums.blurbusters.com/viewtopi ... 510#p52233



Also, why do you think you should be getting ~16ms?

Because that's what other tests I've found on the internet get, here's one for example: https://youtu.be/L42nx6ubpfg

And also, I am basing my tests off of the very first tearing that I see on screen, before I even see the muzzle flash, there's screen tearing that occurs around the gun and hands.

[jorimt]

Because that's what other tests I've found on the internet get, here's one for example: https://youtu.be/L42nx6ubpfg

And also, I am basing my tests off of the very first tearing that I see on screen, before I even see the muzzle flash, there's screen tearing that occurs around the gun and hands.

I was in contact with Chris (aka Battle(non)sense) during each of our original G-SYNC tests several years ago, and he's written a guest article on this site (https://blurbusters.com/network-lag/), so I'm indeed familiar with that source.

To be clear, you're not guaranteed the averaged total you see in his, my, or anyone else's results, as the total will vary depending on the given system, game, test equipment, and methodology. You are however typically guaranteed the same differences between totals of varying scenarios.

It's good to hear you're watching for tearlines before the muzzle flash, but due to there primarily being vertical movement in your tests, you could still absolutely be missing the actual first on-screen reaction here and there, increasing your average readings. That, and to fully validate your current test camera and mouse, you'd have to test that same system and monitor with another (already validated) high speed camera and mouse to determine if your test equipment is (or isn't) adding any delay to the averages.

Either way, just try not to get caught on the totals, as they tend to be relative across different setups. Again, you're looking for the differences between scenarios with the high speed test method, not necessarily your absolute input lag.

EDIT: Duh, I almost forgot to add that I noticed your camera was in portrait orientation, and your led was below and away from your monitor, both of which could affect your averages. Ideally, you want the scanout direction of your phone camera to be oriented with the scanout direction of your monitor (e.g. landscape; specifically turned so that the scanout on your phone camera is going from top to bottom, not left to right or right to left or bottom to top), and you want your led to be affixed to say, the center of either side of your screen, just so that it doesn't occur in a much lower or higher part of the camera scanout than the on-screen reactions that follow (basically, it's best to have the led and monitor as close to the same 2D space as possible when testing like this). Anyway, both of those aspects could be skewing your current results as well, and oh, finally, make sure you put some tape over your test mouse's sensor so it never registers tracking when you click.

[RealNC]

does FSO gives you input lag? and is csgo running in fso also ?

No input lag. As to whether it's active in csgo or in any other game, just change the Windows sound volume through a keyboard shortcut (like the media volume keys) and see if the Windows volume bar appears in the upper left corner of the screen. If yes, then FSO is active.

[1000WATT]

The presence of the volume bar does not mean that the application is operating in FSO mode.
The volume bars will be in both windowed mode and windowed fullscreen.
The absence of volume bars indicates that the application is running in FSE mode or no DWM.
In addition, even if the application was in FSO at the time the volume bars were displayed, the application would be put into composite mode. Until the volume bars disappear.

But maybe I got it wrong again.

[LLm1]

It's a bit worse than mine, I measured mine at about 40ms, and I've got a ryzen 3600 4.2ghz, 2060 super @1650mhz and tridentz ddr4 RAM with xmp enabled.
But I used my keyboard to measure, that generally gives slightly worse results, also my monitor is 120hz not 144hz.

I assume you've optimised your windows right? If yes then I'm not really sure, sometimes people on here say they've got like 20ms input lag, and I'm like wtf, how'd they get that? My best guess is the CPU difference as the ryzen 3000s series has some problems with that, but here you are having a similar problem.

What's your clock speeds on your CPU and GPU btw? The difference between 4GHz and 5Ghz can give you like extra 4ms+, similar situation with GPU too.
Try fixing your clock to 5GHz, if you can, and fix your GPU clock too, maybe your CPU/GPU was boosting down when in CSGO cause that game isn't really that demanding? You can disable hyperthreading to be able to boost clock speed even higher.

Yeah when I used my keyboards reactive lighting it was giving me a reading of 14ms so it's obviously not a viable way of testing. Also what were you using for your camera, and at what fps were you recording at?

[Chief Blur Buster]

EDIT: Duh, I almost forgot to add that I noticed your camera was in portrait orientation, and your led was below and away from your monitor, both of which could affect your averages. Ideally, you want the scanout direction of your phone camera to be oriented with the scanout direction of your monitor (e.g. landscape; specifically turned so that the scanout on your phone camera is going from top to bottom, not left to right or right to left or bottom to top), and you want your led to be affixed to say, the center of either side of your screen, just so that it doesn't occur in a much lower or higher part of the camera scanout than the on-screen reactions that follow (basically, it's best to have the led and monitor as close to the same 2D space as possible when testing like this). Anyway, both of those aspects could be skewing your current results as well, and oh, finally, make sure you put some tape over your test mouse's sensor so it never registers tracking when you click.

This is correct, scanout direction error will add 1 highspeed video frame of error margin -- 1/960sec. Galaxy scans sideways when videoing portrait, so muzzle flash should be directly above the LED, to reduce the error.

To reduce camera-sensor scanout error margin in high speed video camera tests:
For landscape, you want LED + display stimuli at same vertical position ("horizontal" pixel row)
For portrait, you want LED + display stimuli at same horizontal position ("vertical" pixel row)

But, here in this case, it would fudge the number around by only 1ms (tops).

____

Commentary for OP / thread readers:

I've seen lower numbers I've seen in the past, could be mouse button antibounce. But that AOC monitor is known to have low lag, though being only 144 Hz, it will have up to a 1/144sec = 6.9ms first-reaction lag if you're looking at a narrow area (e.g. muzzle flash).

Known latency segments as a reference:

• USB lag = about a millisecond
• Frametime lag = 1/300sec = 3.3ms for GPU to render frame
• Scanout lag (GPU-to-monitor delivery+refresh lag) = 1/144sec = 6.9ms (averages for small-area first-reaction would be about halftime that).
• Monitor lag = approx 4ms VSYNC OFF (checked multiple sites)

I've seen much lower latency in this specific high speed video test using these settings (with 144Hz + 300fps + VSYNC OFF). Could be the usual mouse button antibounce logic, I've seen antibounce processing get that horrible lag before although consistent across 2 mice?

Personally I am a proponent of optical-button mice nowadays. I have a Razer Viper 8KHz and the optical buttons "feels" better/faster to me in Windows, even though I am not a professional player. It doesn't need the antibounce processing.

I suspect a large (significant) latency reduction may occur with an optical-button mouse, if you're measuring click latency.

Also, not recently tested, but be noted that the muzzle flash lag of a 2020 version of CS:GO may be slightly different from 2014 version of CS:GO where we're able to get reliable ballparks of sub-20ms click latencies. More testing to verify 2020 CS:GO behavior may be needed.

Some testers prefer movement lag tests, over click lag tests, both have their pros/cons.

[jorimt]

You could also try the look method I used for CS:GO. My testing methodology and link to the CS:GO custom test map below (watch the video in that section for more context as well):
https://blurbusters.com/gsync/gsync101- ... method-101

It's a far easier method to catch first on-screen tearlines, and It's how I achieved a ~16ms average in your current test scenario:

[deama]

It's a bit worse than mine, I measured mine at about 40ms, and I've got a ryzen 3600 4.2ghz, 2060 super @1650mhz and tridentz ddr4 RAM with xmp enabled.
But I used my keyboard to measure, that generally gives slightly worse results, also my monitor is 120hz not 144hz.

I assume you've optimised your windows right? If yes then I'm not really sure, sometimes people on here say they've got like 20ms input lag, and I'm like wtf, how'd they get that? My best guess is the CPU difference as the ryzen 3000s series has some problems with that, but here you are having a similar problem.

What's your clock speeds on your CPU and GPU btw? The difference between 4GHz and 5Ghz can give you like extra 4ms+, similar situation with GPU too.
Try fixing your clock to 5GHz, if you can, and fix your GPU clock too, maybe your CPU/GPU was boosting down when in CSGO cause that game isn't really that demanding? You can disable hyperthreading to be able to boost clock speed even higher.

Yeah when I used my keyboards reactive lighting it was giving me a reading of 14ms so it's obviously not a viable way of testing. Also what were you using for your camera, and at what fps were you recording at?

I didn't use the keyboard's lights, I instead used an object and hit a key on my keyboard and counted the first frame that made contact with the key, I then counted until it showed on screen the result I was looking for. I also didn't use the muzzle as it was harder for me to tell, instead I used the key that brought up the weapon's shop as it was instantaneous from what I saw, and it affected most of the screen so it'd be easier to spot.

Here's an example, I was doing it on grim dawn here though.
Don't count the first hit as I missed on that one lol; if you can measure precise timings, the frame I started counting on was at 00:00:19.357 because the frame after it hits the key, so one, two, three... and the last frame, which is 8, at 00:00:19.390 because that's when the menu appears on screen.
https://drive.google.com/file/d/159Vepe ... sp=sharing

I used a 240fps camera, so multiply frames by 4.16 to get latency. Oh and I actually remembered badly, I got 33ms here, sometimes I get ~37ms though because I've got BFI enabled.

This test isn't the best though, as you can see from the 3rd attempt, I got like 10 frames, 41.6ms, but if you look at how the bottle hit the key, it was pretty wonky.
At some point I'll get that LDAT stuff so I can do proper measurements, but until then, this is what I got to work with lol.

[LLm1]

It's a bit worse than mine, I measured mine at about 40ms, and I've got a ryzen 3600 4.2ghz, 2060 super @1650mhz and tridentz ddr4 RAM with xmp enabled.
But I used my keyboard to measure, that generally gives slightly worse results, also my monitor is 120hz not 144hz.

I assume you've optimised your windows right? If yes then I'm not really sure, sometimes people on here say they've got like 20ms input lag, and I'm like wtf, how'd they get that? My best guess is the CPU difference as the ryzen 3000s series has some problems with that, but here you are having a similar problem.

What's your clock speeds on your CPU and GPU btw? The difference between 4GHz and 5Ghz can give you like extra 4ms+, similar situation with GPU too.
Try fixing your clock to 5GHz, if you can, and fix your GPU clock too, maybe your CPU/GPU was boosting down when in CSGO cause that game isn't really that demanding? You can disable hyperthreading to be able to boost clock speed even higher.

Yeah when I used my keyboards reactive lighting it was giving me a reading of 14ms so it's obviously not a viable way of testing. Also what were you using for your camera, and at what fps were you recording at?

I didn't use the keyboard's lights, I instead used an object and hit a key on my keyboard and counted the first frame that made contact with the key, I then counted until it showed on screen the result I was looking for. I also didn't use the muzzle as it was harder for me to tell, instead I used the key that brought up the weapon's shop as it was instantaneous from what I saw, and it affected most of the screen so it'd be easier to spot.

Here's an example, I was doing it on grim dawn here though.
Don't count the first hit as I missed on that one lol; if you can measure precise timings, the frame I started counting on was at 00:00:19.357 because the frame after it hits the key, so one, two, three... and the last frame, which is 8, at 00:00:19.390 because that's when the menu appears on screen.
https://drive.google.com/file/d/159Vepe ... sp=sharing

I used a 240fps camera, so multiply frames by 4.16 to get latency. Oh and I actually remembered badly, I got 33ms here, sometimes I get ~37ms though because I've got BFI enabled.

This test isn't the best though, as you can see from the 3rd attempt, I got like 10 frames, 41.6ms, but if you look at how the bottle hit the key, it was pretty wonky.
At some point I'll get that LDAT stuff so I can do proper measurements, but until then, this is what I got to work with lol.

Huh interesting, when I used my camera recording at 480fps I got an average of 14ms. And btw, my windows is optimized and my game was on basically the lowest setting. Vsync and gsync off, true full screen, uncapped fps (~300fps). Also I wonder how much the ldat tool will cost when and if it's released to the public.