40% Less Input Lag Without AMD Anti-Lag or NVidia Ultra-Low Latency 

► Stable Frame-Time / Input-Lag: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-xsXFUVYPIx4.html

DAMN battle(non)sense, you need to do more videos..... your videos are some of the best. As a competitive gamer (not a pro, just decent player) i absolutely love the topics and information you provide.

Doing this test with csgo would be interesting because input lag is so crucial for that game and also because it is so much different than the other games tested (heavily CPU limited, dx9 and very high fps)

I've always capped my FPS in Overwatch to 140, because it feels lower lag than uncapped, despite everyone saying uncapped is better. I never knew if I was imagining it or why it would work like that. But, this seems to confirm that I was right.

Excellent work. Thank you so much to your patreons for funding this! And thank you to you!

It's insane that nobody has discovered this insane flaw yet! Nice job

this video is HUGE value i always use capped fps because i prefer consistent frame times and a cooler gpu. it's amazing to know this is also giving me less lag. capping framerate is something all gamers need to learn about

This was totally unexpected for me and I don't think many are aware of this, not even enthusiasts.

Battle(non)sense we miss you keep the content coming

esport companies should hire this guy to do more testing of this sort.

To achieve the least inputlag you have to have a high amount of FPS even if you have low HZ screen for a better frametime but never letting GPU to reachh 99% usage, when gpu is overloaded it gives a huge input lag, so for example is better to have 120fps with gpu at 80% than 144fps with gpu at 99%

How bizarre.... I wonder how nobody noticed this before?! So strange.

what an awesome test, thank you for taking the time to do this

I would have really liked to see the 5 tests shown at 9:20 repeated with G-Sync active. EDIT: Also perhaps the first and last test repeated with Ultra Low Latency set to On (previously Maximum Pre-rendered Frames = 1) just for posterity.

Oh boy, you must enable reduce buffering in Overwatch. Tested that with G-Sync, V-Sync and G-Sync + V-Sync, and trust me, it reduced lag in every case . You should have that enabled at all time.

Fantastic testing. It echoes my experiences playing playing Battlefront 2 and Apex Legends recently. If I leave the framerate uncapped, both games feel very jittery and uneven. On my 60hz, non-freesync monitor, 90-100 fps seems to worsen responsiveness rather than improve it. However, as soon as I lock to 60 fps using RivaTuner, both suddenly feel really fluid and responsive. My GPU is no longer being tapped out. Having consistent frame times in-line with your monitor refresh rate really is the top priority.

Love your testing. In your tests, you capped your frame rate so basically you are not CPU limited. For the sake of testing, I would love to see the effect of being CPU limited on input lag. Your video has shown that higher FPS does not mean lower input lag, it also depends on your GPU load. I am wondering if it applies to CPU load as well.

You forgot maybe the most important comparison: in-game frame limiter with anti-lag/ULL vs RTSS frame limiter with anti-lag/ULL. RTSS is by far the best solution to cap FPS and get stable frametime, however it is known to add 1 frame extra delay. Since RTSS holds both CPU and GPU to reach 100%, it may as well co-work with anti-lag/ULL and give the ideal maximum. Nice video BTW, thank you!

You do some of the best work in the industry. Bravo man. I wish I could get as much value for the community out of my hardware as you are able to. Fantastic work. Thank you again

Hardware Unboxed just made a video response to this video. Cool stuff!

Chris, if you are reaching high gpu usage that means that the cpu pipeline is faster than gpu at their maximum. The obvious conclusion is that it will pre-render the frame so the latency will increase. That explains why antilag worked. When you fix your fps then you Will run with pre-rendered frames of essencially 0 as both gpu and cpu are abble to render in-time and even sleep. This could be observable using GPUView. What is really interesting is antilag getting higher response time when the expected behaviour is that It should nit affect at all.

Yet another essential video that is changing the way i'm playing. Thank you for bringing those complicated topic to the masses! I'm trying this Low latency mode coupled with Fast V-sync and the results are promising. I can't calculate precisely, it's an overall feeling yet it feels stable and the picture is good. My situation: 4K with 1080ti, it's not good. Really bad tearing, and the input lag is atrocious with Vsync, not mentioning the stutter when going below 60fps, i have a 60hz screen. Well, Fast V-sync works as intented and eliminate tearing while not stuttering if i go below 60fps. What i like the most though, is that stabilized input lag. I didn't realize it was fluctuating that much and it deeply affected my gameplay. Thank you so much battle(non)sense!

Wow. That's awesome. Great video, as Always! I noticed this too when locking the fps in overwatch, it somehow felt snappier than using a higher but variable framerate)( gpu is maxed out). Keep up the good work!

really proud that still are people like you that digs intro the real data and not just opinions. your work there is awesome.

Your RIva Tuner video was a godsend! I was getting stutter that I couldnt figure out how to get rid of it and using Riva and your tips to limit frame rate, eliminated it entirely. Youre videos are incredible and always a pleasure to watch

Now test input lag with new Nvidia In-Driver FPS limiter in 441.87 drivers.

Everything you say it's true and well documented. Input latency is maybe the hardest concept to grasp in gaming because you don't have metrics to measure it. You must understand the math behind it, limit your FPS, and test it yourself. Also it can be counter intuitive, because uncapping the FPS make the game 'feel' more responsive, but in reality that makes that what you are seeing on the screen is not exactly what is happening, as massive input lag is introduced.

This explains why ultra low latency mode made CSGO feel not so smooth to play.

Did you test this in fullscreen exclusive mode? Windows updates have started​ making games run in their own optimised mode (allows for faster alt tabbing but causes​ input lag) when selecting the game to run in fullscreen, in some cases. To fix this you have to go into the properties of the games .exe file and disable fullscreen optimizations, that works for most games (some exceptions) and should put you in fullscreen exclusive mode and reduce input lag.

Hope you do a follow up on this based on possible responses from AMD/Nvidia and factor in using RTSS vs in-game limiters, because my head hurts right now. RTSS limiting with those frametimes feels so smooth though. I must have missed your earlier video. Great work as always!

I wish there was a way to just have the feature automatically turn on every time youre in a scenario with 99% gpu usage and then turn off again when below that, it would make more sense then manually going back and forth.

Sounds like an optimization issue somewhere along the frame processing pipeline, within the GPU driver, DirectX or the game engine itself. Obviously, frame rate limiting will not be an optimal solution since fps can vary wildly depending on gameplay situation. Now I'm pretty curious to hear AMD/Nvidia's explanation.

Danke für dieses video, deine arbeit ist extrem wichtig und hilfreich für eine ganze szene und hat mir sehr geholfen, auch diesmal wieder. mach weiter so!

Some input on the subject (I'm no game engine expert, so take this with a grain of salt): In the past most game would work in the following way: In each frame the game would calculate the logic step that evolved from the previous frame, then it would draw the conclusion of this logical evolution. Let's these steps logical frame and graphical frame. As game engines evolved, some elements which were part of the logical frame moved to the graphical frame, for example simple animations that had little to no effect over the game state besides graphics. Developers realized that you could improve the graphical experience by making the graphical frame independent from the logic one. The logical frame in many situations need to be at a static pace (ex: 60 fps), otherwise some of the collision steps would become too complex. Engines like unity separates them. You can have your game running at 240fps, but the logical frames will be set to 30. This allow very smooth animations and better gpu utilization than if you locked both frames stages. The reason why you must be getting better input lag when reducing the frame rate may be that the logical frames are set to a much lower value then the graphical frame, so by increasing the graphical frame rate you are also increasing the necessary CPU utilization for doing the graphical frames (best case scenario should be only the bureaucracy of sending information to the GPU). So by increasing the graphical frame rate, we may actually be causing a bad influence to the logic frame rate... Games that have both of them locked together should always improve latency when you increase frame rate. Otherwise it becomes complicated. Well... that's my 2 cents.

So a basic summary would be: Don't let your GPU get fully utilized in competative games to reduce input lag. You want to be CPU bound instead of GPU bound basically? Don't use ULL or Anti-Lag if your GPU isn't being fully utilized as it will just increase input lag. So capping CSGO @ 300FPS (Which I roughly maintain about 80-90% of the time.) and using the default option for pre-rendered frames (I believe it's 2 or 3 in GO.) would be the best solution for input lag AND frame time stability, is that right? Or did I miss something. Edit: I remember hearing that capping your FPS also increases your input lag. Have you ever tested the difference in input lag in a game like CSGO if you're getting 500+ fps and capping it at 144, 300, 500 and 999? With the ingame command fps_max.

This is why I searched for your video when I was curious about this. You're the only who truly does in depth testing. Thanks Chris

very intresting topic, it's fascinating how many things there are we don't know about yet

I knew before that keeping your gpu load lower kept input delay lower. I didn’t have any numbers though, but this was preached on blur busters forum for some time. It’s great to have some numbers behind it. Could you test out games that have dx12 and dx11 and run input lag tests where you are gpu bound in both cases? The developer and not the driver have the pre rendered frame control in those games.

Thank you for doing this, I’ve never seen this correlation in a single other report, but I’ve FELT it. I’ve been capping my frame rates for years and things just feel smoother that way, it’s interesting to know the reduced GPU load is a big part of that effect. Even using Gsync, I try to set my in game settings so that they seldom drop below 60 (whatever your refresh rate is), so there’s plenty of “headroom” for the GPU left most of the time. Awesome findings

Surprising discovery, congratulations. So we are wrong to think that when playing the maximum fps we decreased the input lag, but when limiting the fps it would give us lower latencies and better fluency in the game.

That's a sick test! Thanks for this. Hope AMD/NVIDIA will do something, because if I understand it correctly, if you limit your GPU usage and so get less fps, you get a better input lag, and that suck...

I recently discovered this for myself tuning and testing my acer nitro 5 with a gtx 1050. The frametime delay difference of a locked 60 fps 1050 at 80% GPU usage vs unlocked 150+ fps 1080 at 100% GPU usage. It legitimately broken my brain a bit having one so much smoother, while the other so much more blatantly response. Input lag reduction is where it's at, great video.

super interesting results! I always noticed a reduced input lag in OW, when my gpu wasn't maxed out, since I did several tests in the past as well. But I always thought this was an OW related (engine) thing, so thanks a lot for testing this with other games, too!

I went through like 4 videos and this is the only one I understood thank you

So basically...can we get a gpu load limiter? Essentially a variable fps limit so not to max out gpu load but still get max fps at each moment?

Nice finding... I guess next Hyper Ultra Extreme Anti-Lag Latency mode will limit GPU load to 97%...

This guy might be the reason Nvidia Reflex came out in games automatically enabled, it keeps gpu load under 97 % i think

if your gpu usage is less than 99%, you can still be limited by the GPU a certain percentage of the time it's the individual frames which will be limited by the GPU or not, so an "immediate gpu usage" or "average gpu usage in the last second" will not tell you what's up which is exactly why with 80% gpu usage you can gain fps by upgrading your gpu, and with 99% gpu usage you can gain fps by upgrading your cpu, and you can often gain fps by upgrading your ram

This is by far the best video I've watched on this. I've been researching for hours, thank you for this.

I've known about this for a long time. It's mysterious. But about NULL/AntiLag, testing with uncapped framerate should do very little, the render queue only fills up when the CPU outpaces the GPU. Limiting the framerate is not a valid case because then the engine waits and does not prepare a new frame. The worst case is when you are limited by Vsync, the engine will then prepare new frames until the queue is fully filled up. This is the real cause of Vsync input lag. You can then understand why the "58fps trick" works, by making the engine wait instead of filling up the queue you reduce input lag. If you know how most game engines are put together you will know that calling the Swap() function in your graphics API with Vsync enabled will cause the thread to stall until the frame is presented. Since the driver and hardware is a black box and we don't know exactly what happens when you call this function this is where the mystery begins. In order for the render queue to work, that means the driver has to present a false backbuffer so that the engine continues with the next frame instead of stalling. For the render queue to fill up, the CPU framerate has to be higher than the GPU framerate. If Vsync truly capped your framerate to exactly 60 then the queue could not fill up. This presents another mystery, the game engine measures the framerate by using the high precision timer to count the nanoseconds between calls. If say the CPU produces frames faster than 60 to fill up the queue, then the delta time information should also be lower than 16.6ms but this doesn't appear to be the case. Does the driver fudge the timer as well? Now the most mysterious part, which is what you've run into here. Input is taken by the engine before the frame is processed so that it can be used in the next state of the game. Input is taken instantaneously at that moment and does not update until the next frame. The lag should always be at least the time between the frame is processed and it appears on the screen, which is Input + Wait + Swap() + Queue + Sync (even without vsync the monitor still has to cycle) + Pixel Response Time. Capping the framerate should make the engine wait a certain period of time before processing the next frame, that should be extra lag. BUT, when you cap the framerate, you get a lower input lag than if you are limited by hardware. This lower input lag appears to be proportional to your potential maximum framerate. This implies that the frame is actually processed just before it appears on the screen, *even in the case of Vsync with a queue of 1!!!!!* But the thread should be stalling, which should add lag, not reduce it. But the result is you can have input lag comparable to 100fps with a cap of 20fps, but only if you could reach 100fps without the cap. This is what you'll truly have to explain to me, please forward my comment to your contacts.

GPU BOUND GAMES 95 - 99% = TURN ULTRA CPU BOUND GAMES = TURN OFF

Variable framerates and post processing are the biggest input lag culprits. By having a framerate lock (by the engine, not by external progams and vsync) at 144, no post processing effects you will have the best result possible.

Good analysts but incomplete. Perhaps it's best to wait for answers from AMD/Nvidia but more work is needed here. How does Radeon Chill compare to a capped framerate? How about testing a game where the input polling rate isn't asynchronous from the framerate? What is a better gaming experience for competitive gamers? Lower framerates with lower input lag, or perhaps higher framerates with more input lag? Do the lag characteristics change with G/Fsync enabled?. This is a interesting area to keep digging at, please keep up the great work!

Very nice analisys with amazing conclusions. I have always heard it is better to leave FPS free insteado of capping them to reduce lag but it seems it is just the opposite. According to your tests it is better to reduce GPU load by capping FPS. Interesting.

I just have to say that your videos are incredibly interesting and helpful ! Thank you so much for all this helpful information wich would be very hard to get without your channel.

This would beg the question, is there a utility that caps the GPU to a certain usage % rather than fps?

Amazing work. TY for the info. Input lag is one of my favorite issues to tackle when it comes to gaming optimization and you always surprise me with new findings!

Also I recommend reading Blur Buster's G-SYNC 101 article, page 11: In-game vs. External FPS Limiters. Resuming: In-game framerate limiters, being at the game’s engine-level, are or should be almost always free of additional latency. External framerate limiters like RTSS (Riva Tuner Statistics Server), which comes bundled with MSI Afterburner, is a CPU-level FPS limiter, which is the closest an external method can get to the engine-level of an in-game limiter; it introduces 1 frame of delay. Nvidia’s own FPS limiter (accessed with Nvidia Inspector) uses a driver-level FPS limiter, which introduces 2 or more frames of delay. I guess the "new" NVidia Ultra-Low Latency entry and renaming of _maximum pre-rendered frames_ (long live Marketing Departments) is meant to compensate the input lag of it's own driver-level fps limiter and maybe for people who always have V-sync=ON.

You, man, are moving the boundaries of our understanding of gaming. THANK YOU

You sir are a Legend ! Thx for making this Video

ok, i've done some testing with gpuview and came out with a conclusion that even with maximum pre-rendered frames at 1, when gpu-bound, frames are overlapping each other, which increases overall fps, but requires you to get the cpu frame earlier in the pipeline increasing lag, in bfv you have an option called allow future frame rendering disabling which actually prevents that and makes the frames not overlap at all, same like frame-capping, so it reduces lag, but decreases gpu usage and fps as a result

So basically, get your fps as high as you can while staying under 97% gpu load is the best way to go? (Without using either anti-lag or ultra low latency)

There is a new option on the PTR called High Precision Mouse Input and it uses a separate thread for mouse input and I tested it and now my input response time feels like I am on 300 fps even when at a lower cap like 144, it feels so good now

You deserve much praise for this video...great work. How could amd or nvidia not explain this clearer for people? Subbing based on this video.

Very interesting, so a better version of Anti-Lag is something that will dynamically limit the FPS in the game you're playing based on a moving average of your GPU load to something like 95-98% (depending on overhead needed to achieve these results). This could also be something messed up in Windows too and you're essentially manipulating a bug of some kind.

Thanks for the tips man!!! Now I run cs go at 10 fps I can't see shit but I have the fastest reactions ever recorded in the universe thanks fam

TLDR you want stable frametimes, which means you want to cap your framerate with AMD chill + FRTC or whatever Nvidia equivalent at 97% GPU load at all times. In that situation neither CPU nor GPU is under max load, so anti-lag and ultra low latency is not necessary and may even introduce minor latency on top. At least that's how I read/saw this. Thanks Chris :)

This is extremely interesting! Great video as always.

DPC Latency should have a pretty noticeable effect on input lag too. When GPUs are maxed out (or any device actually), DPC latency increases. Also significantly affects network jitter, so minimizing it is critical to reduce input lag.

What about "Low Latency Mode On" ?

CPU utilization of each individual core doesn't really tell you much either. On modern systems, you're unlikely to get into a situation where one CPU thread consistently represents a single app/game thread. Because it is not an efficient way to schedule CPU time. It does happen at times, but it's pretty rare. Usually, even if the game is mostly utilizing say two main threads (which is pretty much the majority of games nowadays, with very few exceptions), these threads are being "thrown around" the actual CPU threads (=cores). As a result, you don't see two "cores" being utilized close to 100%, you see most of your cores being utilized to a certain extent. What makes matters worse is that these CPU monitors are also slow, so they "average out" the values you see. As a result, you can see your cores being utilized to say 20-30% max and *still* be completely CPU limited by the game's main thread. Almost everyone makes this mistake - they see activity on a lot of cores and assume there's no CPU bottleneck and the game is well multithreaded, when in a lot of these cases, the games are still heavily CPU bottlenecked (because their main thread is reaching maximum CPU utilization, but it doesn't show in CPU "core" utilization) and rarely make use of more than 2-3 actual main threads. Even pros such as Digital Foundry consistently make this mistake, which drives me nuts, because not only should they know better, but they're also effectively confirming these misconceptions in their viewers... :( If you want to know whether or not your game is CPU limited and how many threads it actually uses, you need to use a tool that allows you to monitor the CPU utilization of the actual threads of the application/game. Only then you have correct data. Personally, I use Process Explorer for this.

Finally a test that relieve me... I've been wondering for years why cod4 is so frikin smooth compared to other games with the same or close fps values... Now it makes sense that my gpu in cod4 capable to give me 500fps but i limit it to 250 that makes gpu load is about 37-42% But in other demanding games i dont put limit to get 200 fps and the gpu is at 99%.. I was always put the burden on the new games engines.. But now i know the truth uhhh

Getting a better monitor is most important.

You put an incredible amount of effort into this video, great job.

Thanks for this. I had this theory for a while. So I have locked my fps to 120. I'm on a 240hz monitor. It just feels better with hitregs. I9 9900k and 2080 Ti

That makes so much sense! If you reserve some resources of the CPU and GPU to process the inputs more effectively than this results in better latency.

I already feel smarter by listening to his voice.

just found out the low input lag feature on the latest Nvidia drivers, thank you for letting know about the new features :)

don't run the video card at full load, even if the fps is high, there will be high input lag

i knew it, my games always felt far more responsive when locking the frame at 30 compared to letting it run wild around 40-70. Glad to see my hypothesis gets confirmed and that i should disable anti lag to reduce input lag even further

This is with low latency mode entirely disabled versus ultra right? What about just having it 'on' (the old 1 buffer mode)

Very Interesting. Your the 1st to go in-depth about this topic. Everyone else is just turn on and move on. I'm wondering now if I should turn it back to off as I cap my fps to 160 for most games.

How does adaptive sync affect this?

This changes everything. If lower framerates can lead to a better experience, soon, GPU marketing will be less about FPS or frame times, and more focused on input lag and frame time variance.

WELCOME BACK BATTLENONSENSE

Holy nuts i was like "what??, how???" while watching this, it's super interesting, gonna give it a try, thanks dude i'm a sub now

Dat ist Wissenschaft! Danke dafür

This effect has been known for quite a while now. I've been suggesting to people to use frame limiting to lower their input lag in games.

It actually makes sense on hardware level. My assumption : 1.the gpu architecture is similar to the cpu (instruction fetch/decode phases etc.) 2. The need for loading/saving registers happens for each frame. Conclusion : less frames, means less decoding/loading instructions and saving/loading register data, that means less overhead which give you a lower latency. You sacrifice the computing peak performance for a lower latency PS.: if I made a mistake, feel free to tell me

This is eye-opening, thank you for testing them!

Have you ever heard back from AMD and do you know if this issue still persists on modern graphics cards?

I'm surprised that this is new info. One of the perks of owning lower end gpus over the years is noticing these "anomalies" and random performance issues even though fps may be fine. The first time i noticed 99% gpu use causing poor performance was on my Radeon 4890, but that was paired with an old CPU so ..lesser.. performance was expected, and it didn't happen often, which is one of the reasons it became obvious wheenn it did happen, funnily enough. I stopped worrying about it after getting my new system though, but it's much more noticable in the lower fps ranges 50-80 on lower end hardware. Been saying for years I'd love a driver setting to cap gpu performance by %, like maxGpuUse=90%.

Could you please test this with csgo? I heard that this game doesn't like frame cap and I also want to know how the input lag is at like ~300fps

I know this comment might be lost in the list but ill put it anyway. From my testing on my gtx 1080 ive found that default settings works best. Any programme settings changed would add lag and even the full rgb setting. The only change that works best is display scaling (no scaling). The same conclusion as yours apply. Furthermore ive seen that enabling vsync below the refresh rate makes motion more in sync on a gsync monitor (if it makes sense) its easy to verify by shaking the mouse a little from left to right. Hope it helps.

I SWEAR IVE BEEN NOTICING THIS!

It actually makes complete sense. With low latency/anti-lag on you are limiting or completely removing the number of pre-rendered frames that the GPU can prepare. When GPU load is below 100% and these features are disabled, it can use that extra processing time to render the next frame(s) in advance and reducing input lag. Seems like these features are working as intended, just not beneficial for every use case.

This is extremely interesting. Now I get why if I enable the fps limiter I have improved reaction times.

So this is why rocket league started to feel more sluggish when I moved to my 240hz monitor. My gpu was pretty much always 100% whereas before with my 144hz monitor it wasn't. Makes perfect sense now!

i FUCKING love your scientifc methodology. Rare to see in RU-vid video reviews. The results in this video are also a huge revelation. Who could have thought that lower fps can result in lower latency? The math doesn't make sense.

Rtss scanline sync requires the GPU load to stay under around 80% for it to work. I wonder if it has to do with these results.