90% finished input lag in CS2/Valorant - a datapoint

[yee]

Here to provide a data point and hopefully give something back to the Blurbuster community. Everything in this post is subjective and based on my own experiences/testing.

Sorry for the overzealous title but I feel like I've finally reached a system I'm happy with - all there's left on this generation of hardware is to monitor for OC stability and potentially adapt to new patches from games. I'd also like to apologize for lack of formatting as I originallay typed most of this post up on my phone.

A little background: 15k elo in cs2 right now, 3k hours total played in csgo/cs2. Peak ascendant 1 Val with maybe 300 hours. By no means a talented player but consider myself a dedicated tweaker. Probably spent couple hundred hours tweaking.

Specs:
z790 apex, 13700k, 4090 MSI x trio, oloy 8000cl36 16x2, 990 pro, 1000w tuf psu, noctua d15
Former specs:
Alienware prebuilt, 9900k, 2080ti, 3200 kingston ram (no please don't get a prebuilt)

Peripherals:
Alienware 2521h (I've tried the 2566k and dyac is worth it, otherwise the 2521h's IPS is prettier)
Viper v2 pro, wooting 60he, artisan mousepad (it's meh)

I'm going to categorize the perceived input lag impact into small, medium, big and huge. I coin small to be 50% chance that it's my own placebo and big being an difference that could be qualitatively seen over different permutations of settings. Anything marked as huge can constantly shave off 1-2 seconds in the range in valorant, 50 stationary bots. My best time is 43 seconds. For reference, at current settings 2521h's Reflex latency analyzer in valorant range average 5.5ms but I've learned that click latency doesn't translate well to "mousefeel".

System:
CPU oc: huge for cs esp on older system, medium for valorant
HT off: medium for cs
E cores off: big for cs
E cores on: medium for Valorant
Disable C states: big for cs2 at least
Adaptive vs static cpu voltage: small on 9900k in favor of adaptive, 13700k none
4800 to 8000 xmp: big (less microstutters)
8000xmp to 8400cl36: small
8000xmp to cl34 plus tuned secondary timings: none
2521h response time: big, though I still use fast (slowest out of 3) since enemies ferrari peek on the faster settings.
4k polling: big but not exactly input lag. Hard to put in words.
Mouse in USB 3.0 port: small
GPU core +120: none
GPU memory +1500: big
MSI mode on 2080ti: none
MSI mode on 4090: small
Note on Wooting: yeah if you're plat+ in valorant get something with rapid trigger. CS probably not that different.

Windows settings I've always just applied without testing:
Disabling device manager power saves
High performance power plan
Disable full screen optimization
Updating to newest nvidia drivers
Turning off xbox dvr etc
Not installing peripherial applications (Razer synapse, Steelseries gg etc)

Windows:
HAGS on: small
Game mode off: big for cs, none for valorant
Audio enhancements off: medium
Updating Intel management engine and sound card drivers: big (weird one, could be something else)
Bitsum highest performance power plan: none vs high performance
Bitsum performance mode: none
Real-time priority for valorant: big for 9900k, 13700k small.
Real-time priority for CS2: small, regressing to pretty big negative until restarting the game. It's a weird one. I leave it at normal.
Override Application high DPI scaling: none
LLM ultra (with reflex boost on):none for valorant, small for cs
Reflex on and boost: big and medium in both games

Adding exes and setting high performance in windows graphics settings: small
Excluding exes windows defender: none but felt like excluding faceit/esea etc impacted hitreg
Display scaling on GPU: small
NOT ticking override display scaling in nvcp: medium for cs, none for valorant
Disable dynamictick + hpet: big for old system, negative for new system
Win32 priority separation 28: for old system, small, none for new system

Syncs:
VSync off: huge in valorant, big in cs
Gsync off vs on + nvcp capped at 355 vs uncapped on 9900k: none
Nvcp cap vs in valorant in game cap on 9900k: small in favor of nvcp
Sidenote I actually used nvcp cap 355 in addition to in game cap 350 on 9900k system with gsync, vsync off. This gave me the least "floaty" mousefeel.
Advantage of uncapped vs gsync on new system: medium

Valorant:
rawinputbuffer: this I've actually tested with the Reflex latency analyzer, -0.3ms with it on but the main benefit is making high polling rate mouse less "floaty". small
Settings all low (in game 0.1ms gpu) vs cranking up settings to 0.4ms gpu: small. I cranked it up for better visuals.
Multithread rendering on: small. Keeping this one off makes the mouse less "floaty", but fps drops significantly in fights. Feels as if I turned RAM oc off.

CS:
Fps max 0 vs 300-9999: medium in cs2, none in csgo
Cmaa2 over msaax4: small
Shadows, AO, HDR: none on 4090

That's all for now. Will update if I think of anything. If a 14900KS releases I might be able to report back on 9000mhz+ ram.

Please keep in mind this is one data point. Your system WILL be different.

Thanks to Chief, jorimt and others for their excellent work.
Carry on...

Edit 1: Wording, formatting and MSI mode.
Edit 2: Wording to better reflect that 3k hours was played over csgo and cs2 - thanks MegaMelmek


Update:
11/7/2023
Since the last post I've gotten an upgrade to 14900k, swapped my display from Alienware 2521h to Asus PG248QP and my mouse to the Viper Mini SE. Wanted to share my updated settings and thoughts.

- Disabling HT and 5.8 all core nets about 100 fps more on average in CS2 compared to 13700k on similar settings in comp, though I find that leaving the 14900k stock feels more consistant in game.
- I had to clear CS2's shader cache to get rid of stutters upon reverting to stock (as mentioned by Gamer Nexus, I had 7-22 ms spikes on r_show_build_info)
- I did lose 8000mhz ram on the 13700k and could only get 7600cl36 stable with 1.5V RAM VDD/VDDQ + Auto CPU VDDQ/SA on 14900k, though performace hit was not really noticeable.
- For a reference point, on the newly released workshop map "FAST AIM/REFLEX TRAINING [CS2]" I currently average about 1200 fps. r_show_build_info frametime lows can dip to 4-6ms in a heavy 5v5 exec/retake where cl_showfps might show ~500fps. Curious what frametime lows would be on a X3D chip.
- I've also removed -threads 8 (I've also tested -threads 7) since it seems to negatively impact hitreg in game.

- I've enabled Windows 10's game mode since it made my Viper Mini SE on 8k polling rate feel more consistant.
- Viper Mini SE is probably an upgrade if the shape is right - it just "follows my hand" more.

- PG248QP's motion clarity is a huge upgrade from the 2521h, but it's mostly a comfort thing. Mousefeel isn't that much smoother/faster.
- 500hz felt only marginally faster than 360hz on the PG248QP. (540hz only on Windows 11 afaik). This would be a "small" using metrics from the original post. It's probably more impactful in Valorant than CS though, since graphical style differences.
- I also have an Acer X27U OLED and honestly feel that I'd trade my PG248QP for an 24-25 inch 360hz OLED if it existed. So pretty.
- ULMB2 adds input delay and dimmed the display too much. If I had to quantify it I'd say its about 60% of gsync's delay at 500hz. I might have used it if it only dimmed the display or only slowed down my mouse but not both - might as well use OLED. Curious to see real tests on this one cause I have a suspicion it's the lower brightness that translated to perceived input delay.
- PG248QP's OSD bug (glitched graphics, garbled unreadable text) can be fixed by unplugging and replugging the monitor's power cable. I've owned this display for 3 weeks and it's happened 3 times.

Edit 1: added date

[cyberdeath]

Thanks for the post! I run a similar setup: 12700kf, Asus ROG z690 Hero, 32GB Corsair DDR5 @ 6000Hz (my CPU memory controller was not in the good batch so that's all it can handle stable), eVGA GF RTX 3080 Ti, 990 Pro (2x), 1000W Corsair PSU, Noctua D15, and Alienware 2521H (but I use ULMB @ 240Hz).

I was curious about a few of your findings:
* When you mention running CS2 in realtime mode, do you mean with the -high flag or do you mean using something like Process Lasso and setting realtime?
* Do you use the nVIDIA Reflex On within CS2 or just nVIDIA LLM On? Have you measured improvements? When I tested with nVIDIA reflex on my monitor, I didn't see any improvement, so I'm curious what you experienced.
* When you mentioned excluding exe from Defender it impacted hitreg? Positively or negatively?
* I noticed you mentioned you saw negative performance from enabling useplatformtick and disabledynamictick, what kind of difference did you see in CS2?
I have these both enabled:
useplatformtick Yes
disabledynamictick Yes

Again, thanks for sharing!

[yee]

Thanks for the post! I run a similar setup: 12700kf, Asus ROG z690 Hero, 32GB Corsair DDR5 @ 6000Hz (my CPU memory controller was not in the good batch so that's all it can handle stable), eVGA GF RTX 3080 Ti, 990 Pro (2x), 1000W Corsair PSU, Noctua D15, and Alienware 2521H (but I use ULMB @ 240Hz).

I was curious about a few of your findings:
* When you mention running CS2 in realtime mode, do you mean with the -high flag or do you mean using something like Process Lasso and setting realtime?
* Do you use the nVIDIA Reflex On within CS2 or just nVIDIA LLM On? Have you measured improvements? When I tested with nVIDIA reflex on my monitor, I didn't see any improvement, so I'm curious what you experienced.
* When you mentioned excluding exe from Defender it impacted hitreg? Positively or negatively?
* I noticed you mentioned you saw negative performance from enabling useplatformtick and disabledynamictick, what kind of difference did you see in CS2?
I have these both enabled:
useplatformtick Yes
disabledynamictick Yes

Again, thanks for sharing!

Hi there cyberdeath,

In my experience, I've found no difference between using -high flag, setting process priority and using processlasso to set the priority. Though for convenience I've mostly used process lasso.

I've used reflex in both games with LLM on ultra. I remember when it was called "Max-prerendered frames" setting it to zero (equivalent to LLM afaik) would give me microstutters and it was a consensus in the Chinese CSGO community and somewhat on reddit too that it was best set to 1, but at some point, probably couple month after the naming conversion, setting LLM to ultra no longer gave me stutters in CSGO. This was on the 9900k system btw. Been on Ultra ever since.

It's worth mentioning that I'm convinced that reflex supersedes the LLM setting in valorant but in CS2 I'm not so sure. The difference could be smaller than 50% placebo but more datapoints would definitely be helpful, even if it's just "mousefeel" reports from veteran fps players on the 3 LLM settings x the 3 reflex settings.

Sorry if I wasn't clear enough but except where I've said that I chose the slower setting (like 0.4ms gpu render over 0.1ms for visuals in valorant) I've always applied the faster setting. Placebo is hell of a drug.

Excluding applications from Windows defender gave better hitreg. This was most obvious on 5E, basically Faceit for China, on my old system. This was mentioned in a really obscure video on Bilibili and it was actually night and day. I hesitate to recommend it as a blanket setting because YMMV but I'd exclude faceit, esea, exitlag and anything mouse related. Please only do this at your own risk and verify that the excluded application is not malicious or if you're like me use another computer for everything non-gaming

I had the same two settings on the older system plus disabling HPET in device manager. I had faster albeit slightly "floatier" mouse movements and worse hitreg in csgo but I stuck with it. In cs2 I only briefly tested it and it didn't give me "faster" mouse (sorry for the lack of a better descriptor but I think a veteran cs player would probably understand, faster being opposite to swampy) so I didn't bother testing further. Too many variables to test in cs2 so I had to cut down on number of enumerations. If you have success with the bcdboot suite of settings in cs2 please do let me know. In Valorant though it definitely gave me worse hitreg (on 10ping) even when only disabledynamictick is applied so it might not be so innocent as a "power saving feature". Perhaps riot accounted for it in their netcode? Who knows.

Hopefully I've answered your questions.
Happy tweaking!

[MegaMelmek]

3000 hours on cs2 thats 125 days so hard to belive that….
If some one spels word placebo he dont know whats real input lag is…
You probaly talking about sytem latency 1ms there, 5ms there and so on. For average lvl 10 thats not latency….

[Chief Blur Buster]

3000 hours on cs2 thats 125 days so hard to belive that….
If some one spels word placebo he dont know whats real input lag is…
You probaly talking about sytem latency 1ms there, 5ms there and so on. For average lvl 10 thats not latency….

3000 hours does happen often for career esports players (dozens of who post here) or bigtime hobby players who don't work / or work from home / or trying to make a living via streaming.

I've seen esports players spend that many hours in just one year (literally like a fulltime job plus reasonable overtime = 3000 hours total), and Valorant is way more than a year old now.

There are esports players who specializes only in one game, such as only CS:GO, or only Valorant, etc.

There are million-subscriber youtubers who also post here too (albiet under more anonymous usernames).

Now that said, the raising of "placebo" is also a point of contention -- one person' placebo may not be another persons' placebo, since there literally infinite number of causes of lag of varying types. Lag is also not a single number (since displays don't refresh all pixels simultaneously) -- there's other lag metrics like network lag, lag compensation, consistency, volatility, lag differentials for different parts of screen, and literially ZILLONS of lag-related complaints that sometimes are unrelated to each other -- and YES -- "placebo" is included, but yelling "placebo" is frowned on around here like yelling "fire" in a theater -- and sometimes forum rules are enforced on "Always Respect Newbies" here over a "placebo" argument.

[MegaMelmek]

When i read i try to focus on detail and take most from it but when someone claim this:

A little background: 15k elo in cs2 right now, 3k hours played.

there is for sure something wrong and is hard to belive rest of the writen text

[yee]

Oh no sorry for the confusion. I meant 3k hours in csgo since steam tallies everything together under "hours played". That's my bad and thanks for pointing it out. 15k cs2 elo and ascendant also means only about 90% percentile in those games.

I put my hours to say that I'm not a professional player or hardcore streamer and yet I'd like to post my findings because I think these data points have their merit nonetheless.

To elaborate about system latency vs "mousefeel", I feel like the reflex latency analyzer (and I assume, LDAT) gives "digital" reading while player reports of "mousefeel", however affected by placebo effects, are "analog". Some things are captured well by digital measurements yet our mouse polls at 1-8k Hz which feeds into a CPU running at several GHz that interrupts and waits. I have no doubt a pro player would on average be more perceptive to changes in system latency yet AFAIK many do not optimize their system. It's also been proven time after time even a "fps-naive" user can discern system latency fairly well on a high-end, high refresh rate system.

Again, I had no intention of misleading anyone into thinking I have 3k hours in cs2. As someone who's only had access for maybe a month I didn't even think that's possible. Still I'd encourage everyone to share their findings since latency tests doesn't show dynamic changes in input latency and anecdotal reports, however riddled with placebo, are still the best we have until enough data is gathered to compile a trend.

And again, thanks Chief - if you'd like to edit my post to say 3k hours in csgo please feel free to do so.

[yee]

Now that said, the raising of "placebo" is also a point of contention -- one person' placebo may not be another persons' placebo, since there literally infinite number of causes of lag of varying types. Lag is also not a single number (since displays don't refresh all pixels simultaneously) -- there's other lag metrics like network lag, lag compensation, consistency, volatility, lag differentials for different parts of screen, and literially ZILLONS of lag-related complaints that sometimes are unrelated to each other -- and YES -- "placebo" is included, but yelling "placebo" is frowned on around here like yelling "fire" in a theater -- and sometimes forum rules are enforced on "Always Respect Newbies" here over a "placebo" argument.

I'd also like to clarify that I use the word "placebo" because the smaller differences, at least to me, are so subtle that I can't say for sure that it was "this particular change" that enabled me to frag better/had better mouse movement/x seconds faster. I definitely think every change to a system however minor has an effect and the "50%" placebo applies only to myself because I can't be 100% sure about my own findings.

Again, to be clear, I had no intention to say that one particular setting is 50% placebo, or had a however small, medium, big or huge impact on everyone's system - I just didn't know how to word it better.

I'll refrain from using the word "placebo" and try to be more precise from now on.

Thanks for giving me the benefit of the doubt.

[Chief Blur Buster]

I'll refrain from using the word "placebo" and try to be more precise from now on.

Thanks for giving me the benefit of the doubt.

Thanks.

We have had situations where people yelled "placebo" and then a lag tester (e.g. OSRTT) found real problem, or some more custom software caught occasional transient lag spikes (e.g. 50ms frametime spikes once a second) caused by background RGB-animation software, or many other obscure problems (like one of the millions of background apps).... That is missed by most lag testing software!

And we've had situations where there is a genuine placebo. But, we've had so many placebo's proven non-placebos, that the word "placebo" is getting verboten around here.

You're good and appreciate the acknowledge;

...it's just that (in the past) I've seen some really fierce forum escalations over the placebo debate in the past (full of 4-letter words and/or anger at each other) that led to forum bans before, that's what I try to prevent. You never went that far, so don't worry.

Unless a forum member has literally over $1 million dollars worth of testing equipment and over 100+ science papers -- it's very difficult to have a true perfectly genuine "placebo truth pass" around here, to correctly use the word "placebo".

This is why yelling "placebo" around here is exactly the same as yelling "fire" in a theater, around these Blur Busters premises.

[imprecise]

Peripherals:
Alienware 2521h (I've tried the 2566k and dyac is worth it, otherwise the 2521h's IPS is prettier)
Viper v2 pro, wooting 60he, artisan mousepad (it's meh)

Thanks for your work on this post!

I just want to point out Alienware devices will stealth install AWCC during windows updates, and can decrease performance. This includes mouse, keyboard, monitor, and probably other devices. The main issue I found with it is interference with the power profiles.

https://www.tomshardware.com/news/how-t ... es-windows

https://www.reddit.com/r/Alienware/comm ... _installs/