People with input lag; What GPU do you have?

[inputlagenjoyer666]

Hello fellow input laggers, I've had input lag for 5-6 years now, changed pc parts sporadically, etc, problem still persistent.
Over time I've bought new monitors with increasing refresh rate, bought stronger cpu & gpus, pushed my max framerate in game (CSGO/CS2) even higher...

The latency has been reduced, but the persisent low hz feeling, floaty mouse, variable input lag has always been persistent. Something I've thought about is the fact that for CSGO/CS2/Valorant, my games have always been GPU bound when I uncap my FPS. I've been capping the FPS to a value the game can hold reasonably consistently, but during complex graphical/computational situations (lets say HE grenade blowing up a smoke in CS2 or multiple ults/abilities on a valorant execute round) i still think my GPU is peaking out in rendering situations (multirendering maybe?) for very tiny amounts of time, so small that it might not show as 100% but some value close to it (i see 80-95% sometimes in software like nv frameview, msi afterburner, e.t.c.) possibly altering the render queue, or making the game microstutter, etc?

I know nvidia reflex exists, but that still allows render queues to exist, just that they are updated dynamically, and LLM (lowlatencymode) in nvidia control panel isn't always honored by game engines, according to what i read online at least (and it certainly feels that way in game)

My question to you; do any of you have a strong graphics card, lets say RTX 5070 ti, RTX xx80, RTX xx90 series GPU with a capped framerate (or uncapped if you have a really strong GPU such that gpu bound scenarios are basically impossible in your game (cs2/valo) with your specific graphical settings) such that your system is reasonably never GPU bottlenecked AND you suffer from input lag problems like variable input delay?

For reference/anyone curious; I have a 3060 ti w 7800x3d, fps max 400, 360hz monitor. Fps drops in complex situations to like 330 at max but GPU never shows 100%.

[kyube]

Hello fellow input laggers, I've had input lag for 5-6 years now, changed pc parts sporadically, etc, problem still persistent.
Over time I've bought new monitors with increasing refresh rate, bought stronger cpu & gpus, pushed my max framerate in game (CSGO/CS2) even higher...

The latency has been reduced, but the persisent low hz feeling, floaty mouse, variable input lag has always been persistent Something I've thought about is the fact that for CSGO/CS2/Valorant, my games have always been GPU bound when I uncap my FPS. I've been capping the FPS to a value the game can hold reasonably consistently, but during complex graphical/computational situations (lets say HE grenade blowing up a smoke in CS2 or multiple ults/abilities on a valorant execute round) i still think my GPU is peaking out in rendering situations (multirendering maybe?) for very tiny amounts of time, so small that it might not show as 100% but some value close to it (i see 80-95% sometimes in software like nv frameview, msi afterburner, e.t.c.) possibly altering the render queue, or making the game microstutter, etc?.

I know nvidia reflex exists, but that still allows render queues to exist, just that they are updated dynamically, and LLM (lowlatencymode) in nvidia control panel isn't always honored by game engines, according to what i read online at least (and it certainly feels that way in game)

My question to you; do any of you have a strong graphics card, lets say RTX 5070 ti, RTX xx80, RTX xx90 series GPU with a capped framerate (or uncapped if you have a really strong GPU such that gpu bound scenarios are basically impossible in your game (cs2/valo) with your specific graphical settings) such that your system is reasonably never GPU bottlenecked AND you suffer from input lag problems like variable input delay?

For reference/anyone curious; I have a 3060 ti w 7800x3d, fps max 400, 360hz monitor. Fps drops in complex situations to like 330 at max but GPU never shows 100%.

You've precisely hit a major point in the chase of low latency.
Purchasing great HW is the #1 "tweak" one can do to "fix" these... redundant and trivial issues people here like to purport.
This is why having 20-35% GPU usage is optimal, as per:
https://www.youtube.com/watch?v=dPMHEyz38TM
https://www.youtube.com/watch?v=8ZRuFaFZh5M
Of course, there's the whole discussion of GPUTime (& in extension, GPUBusy & GPUWait), which would be a more accurate way to evaluate this.

Refer to other threads where I've discussed this in detail.
e.g.:
viewtopic.php?f=10&t=14578
viewtopic.php?f=10&t=14673
viewtopic.php?f=10&t=14642
viewtopic.php?f=10&t=14672
viewtopic.php?f=10&t=14578
viewtopic.php?f=10&t=14652

The main thing to keep in mind is — All of these "issues" are measurable by Windows (.ETL examination with WPA/MXA/xperf/xtw and PresentMon) & external equipment (oscilloscopes + probes, >1000FPS cameras etc.) and can be easily solved

[inputlagenjoyer666]

Thanks for the reply Kyube,

I've read through the linked threads you provided to gain some knowledge. I understand the systems are similiar to mine with regards to the CPU (mostly 7800x3d and a 9800x3d, think i saw one intel as well). I also noticed that these setups all had significantly stronger GPU's than mine - and still report performance issues similiar to my own. While this certainly doesn't disprove my idea of the GPU being the scapegoat of my input latency issues, it does put some doubt whether my GPU really is the main culprit... even though I do understand that getting a better GPU would mean immediate significant latency reduction. Especially considering the fact that I tried playing insanely low resolutions like 800x600 with 360/400 max fps and have the same "low hz" and variable input latency feeling. (it did however slightly reduce the issue, as you might expect).

I noticed that some threads were mentioning microstuttering but really (I think) they were refering to macrostuttering. I don't have an issue with that, they do occur but relatively rarely. Standing in smokes, rendering molotovs on my screen, loud noises like weapon spray & he, all make my game feel more "low hz" and the input latency is very noticeably changing. And its my assumption that this could be due to microstuttering and possibly render queuing, even though I haven't verified with data that this is happening. (I don't know how to, to a respectable level). It is also my understanding that the issue is not directly correlated with FPS increase/decrease, as i noticed the same lowhz/input lag issues with the 800x600 resolution configuration where my system was able to sit at 360 FPS 95% of the time, with occasional slight drops to ~357, and slightly larger drops in complex graphical scenes/situations according to the Windows Game Bar FPS graph (Maybe a flawed measuring tool). However, when FPS drops do happen, the issues are of course amplified.

You mentioned the fact that these issues are measurable - I agree - and I'm interested in trying to solve my issue in this way. I don't have equipment or ability to acquire equipment capable of externally measuring the issue. As such I'm more interested in looking at the data, and based on your replies, is it your opinion that it might be good to start with analyzing xtw results? ]I would be eternally grateful if could provide me with some guidelines as to how I can troubleshoot the issue in this way. If you would want to assist me solving the issue, I can provide any details you'd like, for instance, system components, ram timings, peripherals, anecdotal experience, previous troubleshooting, e.t.c.

EXPO enabled/disabled unfortunately doesn't fix the issue. The mouse feels different however, but very subtly and potentially just placebo.

Interesting sidenote: DanGamesTwitch noted this in his most recent reply in the corresponding thread:

When minimizing and restoring the game, it suddenly becomes much smoother and more lively. This lasts for about 10 seconds, then it becomes choppy again

. I have noticed the exact same issue, for many years, both in CSGO/CS2, and is one of the many ways I know what the refresh rate smoothness & latency could look like, compared to what it feels like most of the time... Coincidence? A general issue, maybe.

[kyube]

I noticed that some threads were mentioning microstuttering but really (I think) they were refering to macrostuttering. I don't have an issue with that, they do occur but relatively rarely. Standing in smokes, rendering molotovs on my screen, loud noises like weapon spray & he, all make my game feel more "low hz" and the input latency is very noticeably changing. And its my assumption that this could be due to microstuttering and possibly render queuing, even though I haven't verified with data that this is happening. (I don't know how to, to a respectable level). It is also my understanding that the issue is not directly correlated with FPS increase/decrease, as i noticed the same lowhz/input lag issues with the 800x600 resolution configuration where my system was able to sit at 360 FPS 95% of the time, with occasional slight drops to ~357, and slightly larger drops in complex graphical scenes/situations according to the Windows Game Bar FPS graph (Maybe a flawed measuring tool). However, when FPS drops do happen, the issues are of course amplified.

When evaluating CS2's performance in particular, one can notice a lot of complaints online about something they refer to as "1% & 0.1% lows"
This is where the “low Hz” feel stems from, because the frame pacing is not consistent.
This is why frame rate is a somewhat deceiving metric to use, as it is essentially latency which we're actually experiencing (360FPS = ~2,77778ms latency)
The best way to give your brain a fundamental, basic representation of how content is supposed to look like is to play a 240FPS video @ 240Hz
That is how every game should run all the time, if the frame pacing is done correctly.
Esports games, however, are very far from such perfection...
The closest thing to that were Reflex Arena & Diabotical iirc.

This is why, as as #1 troubleshooting step, it's always best to evaluate using PresentMon metrics such as FrameTime (also referred to as CPUTime), GPUTime (composed of GPUBusy & GPUWait etc.) & others
Sources:
https://github.com/GameTechDev/PresentMon
https://gamersnexus.net/gpus-cpus-deep- ... discussion

HW choices (UEFI settings, motherboard controllers) & SW choices (background user app SW, Windows settings) noticably affect the aforementioned metrics.
The more CPU Cycles (e.g.: taken on context switches) a user app sw thread (or process) consumes, the less resources your CPU has left to work with on the game thread itself.

You mentioned the fact that these issues are measurable - I agree - and I'm interested in trying to solve my issue in this way. I don't have equipment or ability to acquire equipment capable of externally measuring the issue. As such I'm more interested in looking at the data, and based on your replies, is it your opinion that it might be good to start with analyzing xtw results? ]I would be eternally grateful if could provide me with some guidelines as to how I can troubleshoot the issue in this way. If you would want to assist me solving the issue, I can provide any details you'd like, for instance, system components, ram timings, peripherals, anecdotal experience, previous troubleshooting, e.t.c.

EXPO enabled/disabled unfortunately doesn't fix the issue. The mouse feels different however, but very subtly and potentially just placebo.

It'd help the troubleshooting process if you could include your entire HW & SW stack. A Zentimings screenshot can help too
Once that is known, I can tell you what you could try to troubleshoot

Interesting sidenote: DanGamesTwitch noted this in his most recent reply in the corresponding thread:

When minimizing and restoring the game, it suddenly becomes much smoother and more lively. This lasts for about 10 seconds, then it becomes choppy again

. I have noticed the exact same issue, for many years, both in CSGO/CS2, and is one of the many ways I know what the refresh rate smoothness & latency could look like, compared to what it feels like most of the time... Coincidence? A general issue, maybe.

One can attribute such behavior to improper presentation model choice

Information in regards to that:

phpBB [video]

Since we don't have the ability to use Special-K in CS2 during online play (as it injects into the game and MM is terrible), we must rely on PresentMon to evaluate or app sw tools such as Nsight Graphics or GPUView
Ideally in a repeatable envinronment, such as the CS2 Benchmark map (though I haven't tested how repeatable it is and whether one can extrapolate based on it)

[Kal]

Would you be able to help me as well? I have the same kind of issue, or even worse right now. I'm on a 9800x3d with a 5070.
I can post whatever you need in like a week when my exams are finished.

[inputlagenjoyer666]

Sorry for some days delay, similiar to the previous users reply I also happen to have exams currently which I will be done with by the end of next week. I have invested more time than I should the past few days into trying simple deduction-logic troubleshooting, changing 1 variable such as a peripheral, a software, game settings, recommendations from blur busters, e.t.c, 1 by 1. Unfortunately I don't think I will be able to provide updates very frequently in the coming days due to aforementioned circumstance. I understand if your interest in this thread might fade as a consequence, and I accept that.

Anyways, today I started having a really smooth feeling ingame, similiar to vsync on my typical non-smooth game day, and nonvariable input lag in high complexity scenes in cs2 after I disabled MPO:s, in accordance with your recommendation in one of the threads. I disabled MPO:s with this regedit:

Code: Select all

[HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\Dwm]
"OverlayTestMode"=dword:00000005

However, I read on multiple places online that this setting is broken on Windows 11 as of recent updates (for instance on this forum: https://www.techpowerup.com/forums/thre ... 25.331858/). As such I'm a little worried that it might be a coincidence that the game felt so good, since that does happen on my system sometimes, for 1-2 matches. I will not change any settings and see if the setting stands the test of time in the following days. If this stays the same way, and I get a stronger GPU, I'm confident my issue could be deemed as "resolved". But I don't want to get ahead of myself too much, I've felt this way about different changes many times before.

Should also mention that I switched my inet connection medium from LAN cable to Wi-Fi in close time proximity, just some hours before, to the MPO change. I did however not notice any immediate effects or changes on my game by doing this when i tried playing, but I'm currently still on WiFi when the game was feeling good, so I feel like I can't completely rule it out. (Reason for the change: Paranoid that there might be something bad with my ethernet controller, or ISP, or the combination, since I notice my problems are worse in online 5v5 than offline 5v5 vs bots, even if I throw a lot of util around and stuff). I do however also understand that online 5v5 has higher cpu/gpu overhead due to more complex networking and that it might simply be natural performance difference.

Im gonna provide you with the requested information even though my problem is temporarily at least heavily reduced, since I don't trust it to stay that way:

It'd help the troubleshooting process if you could include your entire HW & SW stack. A Zentimings screenshot can help too
Once that is known, I can tell you what you could try to troubleshoot

SW: Win 11 24H2, Steam w HW accel off, CS2, Discord w HW accel off/Teamspeak 3. No RTSS/Afterburner
Everything default in Windows except: HAGS: disabled, Win32PrioritySeparation=40.
All updates installed & optional drivers provided by Windows update. Also installed newest LAN, WiFi and Chipset driver from mobo website. Currently NVCP LLM set to ON, reflex disabled, In-game frame limiter set to 400 (not NV driver limiter). GPU driver 560.94, provided by Windows upon OS install.

Balanced power plan with following modification: "USB Selective suspending: Disabled". I have messed around a lot with power plan settings using PowerSettingExplorer but usually just end up with massive CPU spiking ingame.

Note that I have tried a billion things in the past like DDU uninstall/install NV drivers, debloated NV drivers, RTSS frame limiter, Process lasso changes like disable core 0 and priority switches on SW, interrupt affinity masks, changing system timer clocks, etc etc... I understand some of these things could be good for performance, but in my experience so far its usually at the cost of something else, or just placebo, and as such I've become quite disillusioned with many of these "fixes" or "improvements", rather. Especially seeing as T1 pros probably aren't using 99% of these things anyway. I know a lot of high rated players in my country who has had their PC's "optimized" by "PC optimization gurus" just to reset their computer some days later, because they felt like their game became cursed, even if it had more throughput in Avg FPS, or something...

HW:
PC: CPU: 7800x3d, GPU: 3060 ti, RAM: G.Skill 32GB (2x16GB) DDR5 6000MHz CL30 Flare X, MOBO: ASUS ROG Strix B650E-F Gaming WIFI, PSU: Corsair RM850X 850W, CPU COOLER: Thermalright Phantom Spirit 120, Monitor: Zowie Xl2566K 360hz

Bios modifications:
EXPO: off, SVM Mode: Off, SMT: off, C-states: off, DF C-states: off, PBO: off, Medium Boostit: On, Spread Spectrum: Off, CPU PCIe ASPM Mode Control: Disabled, Native ASPM: Disabled.

Peripherals: Mouse: Endgame OP1 8k (set on 1000hz), Keyboard: Wooting 60HE, Headset: Logitech G PRO X Headset wired. No device connected to the chipset, Mouse on single ported usb controller 2.0 and Kbd&Headset on ASMedia controller.

Ram timings:

[babooz99]

i have 3060ti 7800x3d as well

[inputlagenjoyer666]

i have 3060ti 7800x3d as well

And judging by your reply history, seems like you have similiar issues. I see you've talked about desync, I thought I had it too, but for me I think thats just what it seems like (but not actually happening) due to frametime variation, microstuttering, maybe causing per frame view angles and bullet spray mismatch vs the server. Also the general low smoothness making people seem like they are ferrari peeking cause their model is like 60hz and at 220-250 ingame velocity (cs2). I see you play valorant, I have exactly the same issues in that game, but the ferrari peeking is obviously not as exxagerated since you move significantly slower in that game.

Babooz, do you also have times, seemingly at random, where the game feels ultrasmooth & "normal"? And then it goes back, without any changes, from one minute to the next? Did you have input/desync with the 1080ti too?

Tried my game today on retake server and it still has a "out of placebo zone" increased smoothness, and my mouse a bit "disconnected" from the FPS drops, which is completely opposite to how it usually is. Maybe try the MPO regedit if you haven't already?

[kyube]

Anyways, today I started having a really smooth feeling ingame, similiar to vsync on my typical non-smooth game day, and nonvariable input lag in high complexity scenes in cs2 after I disabled MPO:s, in accordance with your recommendation in one of the threads. I disabled MPO:s with this regedit:

regedit code

Tinkering with MPO's should be proceeded with caution, as not every game behaves well when you disable or enable them.
I haven't checked whether CS2 employs bitblt or flip model, whose behavior can be estimated here: https://wiki.special-k.info/Presentation_Model
MPO compatibility heavily depends on:
GPU Architecture & vendor, GPU driver, game, Windows version, misc Windows registry changes etc.

Depending on who you ask, one party will tell you to disable MPO always while the other will tell you to enable it.
A lot more context is required before jumping to a "unified" conclusion that applies for all types of workloads, especially in the current software sphere.
There's a lot of variables to nail right, hence why PresentMon and .ETL data are required for benchmarking purposes.
Only then can one find out which one is the best for his use-case.

However, I read on multiple places online that this setting is broken on Windows 11 as of recent updates (for instance on this forum: https://www.techpowerup.com/forums/thre ... 25.331858/). As such I'm a little worried that it might be a coincidence that the game felt so good, since that does happen on my system sometimes, for 1-2 matches. I will not change any settings and see if the setting stands the test of time in the following days. If this stays the same way, and I get a stronger GPU, I'm confident my issue could be deemed as "resolved". But I don't want to get ahead of myself too much, I've felt this way about different changes many times before.

Should also mention that I switched my inet connection medium from LAN cable to Wi-Fi in close time proximity, just some hours before, to the MPO change. I did however not notice any immediate effects or changes on my game by doing this when i tried playing, but I'm currently still on WiFi when the game was feeling good, so I feel like I can't completely rule it out. (Reason for the change: Paranoid that there might be something bad with my ethernet controller, or ISP, or the combination, since I notice my problems are worse in online 5v5 than offline 5v5 vs bots, even if I throw a lot of util around and stuff). I do however also understand that online 5v5 has higher cpu/gpu overhead due to more complex networking and that it might simply be natural performance difference.

Im gonna provide you with the requested information even though my problem is temporarily at least heavily reduced, since I don't trust it to stay that way:

SW: Win 11 24H2, Steam w HW accel off, CS2, Discord w HW accel off/Teamspeak 3. No RTSS/Afterburner
Everything default in Windows except: HAGS: disabled, Win32PrioritySeparation=40.
All updates installed & optional drivers provided by Windows update.
Also installed newest LAN, WiFi and Chipset driver from mobo website.
Currently NVCP LLM set to ON, reflex disabled,
In-game frame limiter set to 400 (not NV driver limiter).
GPU driver 560.94, provided by Windows upon OS install.

Balanced power plan with following modification: "USB Selective suspending: Disabled". I have messed around a lot with power plan settings using PowerSettingExplorer but usually just end up with massive CPU spiking ingame.

I would recommend that you test out W11 23H2, W11 22H2 or W10 22H2 instead.
Run the .reg file found here as a baseline after installation.
I personally believe Windows Defender & Windows Updates are a severe hog to CPU cycles and should be disabled, though you might come to a different conclusion.

As for the driver topic:
I would recommend against using any W11-based drivers (which I assume Windows has downloaded for you ootb)
Your 3060ti got support since the 457.51 – latest driver.
CS2 got a "game ready driver" since 537.42
You have the following branches of driver to test & evaluate which one is the best for your GPU's architecture & CS2:
R535, R545, R550, R555, R560, R565

I would not go past R570 due to the seldom instability reports on the support thread on Reddit. Though, this may be completely false in regards to CS2.
I cannot give you a concrete driver recommendation (reddit ones are a meme), as I lack data in that particular game.
Driver version experience also heavily depends on the version of Windows you're running on, so using reddit recommendations (537.58, 551.23, 560.94, 566.36) is not advised unless benchmarked beforehand.

To shortly touch on networking:
You could experiment with 1GbE or 10GbE NICs based on Intel or Marvell, if you think that this variable may be causing you issues.
Networking hardware (routers, switches, cables) can impact your experience, but I cannot say for certain that this might be the case for you.
There are variables outside of your scope of changing (e.g.: server routing is limited by ISP), other than changing your geographic location.

Note that I have tried a billion things in the past like DDU uninstall/install NV drivers, debloated NV drivers, RTSS frame limiter, Process lasso changes like disable core 0 and priority switches on SW, interrupt affinity masks, changing system timer clocks, etc etc... I understand some of these things could be good for performance, but in my experience so far its usually at the cost of something else, or just placebo, and as such I've become quite disillusioned with many of these "fixes" or "improvements", rather. Especially seeing as T1 pros probably aren't using 99% of these things anyway. I know a lot of high rated players in my country who has had their PC's "optimized" by "PC optimization gurus" just to reset their computer some days later, because they felt like their game became cursed, even if it had more throughput in Avg FPS, or something...

Never trust what pro players do, they don't know what's objectively good.
Their "pop off" moments can be attributed to server routing differences, better set-up systems due to IMC lottery etc.
Most "twitter tweakers" are laymen, who are hired by tournament organizers and destroy PC performance (because they don't understand that "IMC lottery" exists), potentially leading to a severe reduction in a player's capabilities due to microstuttering and/or macrostuttering.
I'm sure most of us have seen a few "oh his game crashed" moments while watching a tournament. That is precisely the cause of those crashes, improperly set-up systems.
It has nothing to do with the game. Almost never had, never will.

HW:
PC: CPU: 7800x3d, GPU: 3060 ti, RAM: G.Skill 32GB (2x16GB) DDR5 6000MHz CL30 Flare X, MOBO: ASUS ROG Strix B650E-F Gaming WIFI, PSU: Corsair RM850X 850W, CPU COOLER: Thermalright Phantom Spirit 120, Monitor: Zowie Xl2566K 360hz

Bios modifications:
EXPO: off, SVM Mode: Off, SMT: off, C-states: off, DF C-states: off, PBO: off, Medium Boostit: On, Spread Spectrum: Off, CPU PCIe ASPM Mode Control: Disabled, Native ASPM: Disabled.

Peripherals: Mouse: Endgame OP1 8k (set on 1000hz), Keyboard: Wooting 60HE, Headset: Logitech G PRO X Headset wired. No device connected to the chipset, Mouse on single ported usb controller 2.0 and Kbd&Headset on ASMedia controller.

Ram timings: image

As for your hardware
The only thing which caught my attention is the mobo choice (due to audio & NIC) & disabling C-states.
As per AMD's server tuning guides, they advise against disabling C-states in BIOS and instead rely on software methods (e.g.: by using Idle disabled)
Whether the latter part (changing Idle mode) is worthwhile for CS2 is debatable & heavily depends on your cooling performance.
The "C-state always enabled" recommendation may also be incorrect, as implementation may differ across mobo vendors.
It's good that you're troubleshooting with EXPO disabled to remove one of the most crucial variables most users battle with.

A few considerations for peripherals, to lower total system latency:
Consider purchasing either a HDMI splitter (best) or a PCIe audio card (slightly worse) to avoid the USB interrupt & latency overhead of the ALC4080 controller (check the audio latency discussion thread, linked in my signature)
CM8828-based or CM9982a-based should net you a substantial (-50ms, if methodology is to be trusted) latency reduction.

As for your peripherals:
The XL2566K has a bit of added processing delay as per some reviewers. Whether this is due to specific firmware or some OSD setting is unknown. Enabling DyAc may increase the added latency, but also severely decrease eye-tracked motion blur, you can experiment with that if you'd like.

This is all I can think of at the time of writing.

[Hyote]

6900 XT and I find it as legendary as the 1080 ti.