Awesome stuff. I think lag analysis videos on a per game basis would work great - at least I would love to see your indepth take on games like Rainbow Six Siege and Valorant. I'd also love to see RTSS included.
found your channel few days ago and im binging every single video they are just so well made there is no one who does this type of stuff better than you. hope some day you get the recognition u deserve
@@noel4140 Yep! Cap to 190, it should slightly reduce input latency and make the image a whole lot more stable. Try it yourself using RTSS, it should be very apparent, particularly if you go back and forth from 190 to 240 or higher. Theres a reason why most pros keep a 190 cap.
Another detailed and informative video. Good job. So, in terms of input lag, the best combination for players who don't use blur reduction is: - Highest possible framerate - GPU utilization under 100% - Nvidia Reflex (or ULLM) And for players using blur reduction (backlight strobing, CRT monitors): - Framerate always above refreshrate - RTSS Scanline Sync with tearline near the bottom of the screen - Fast Vsync on
FYI - Quake Champions PTS supports Raw Input and Nvidia Reflex + Boost now. It'd be pretty cool to see another input lag analysis when the patch hits live servers.
Thanks for the tip! I just re-ran all the same measurements I did for Destiny 2 on QC PTS; I was planning on doing another video specifically about Quake Champions, but... Reflex in QC is literally doing nothing right now. On, Off, On + Boost... makes no difference. In a GPU limited scenario at 60 FPS, all three of those modes return the same lag of around 57 ms. I'm willing to cut them a little slack since they haven't introduced Reflex into the main game yet, but it's a bit annoying that they have an option that doesn't work. Although, I do remember there being a period of... maybe months where the V-Sync option in the settings menu also didn't work. Interesting, though, using raw input for the mouse makes lag worse. Setting the mouse back to DirectX input dropped the lag from 57 ms to 44 ms. I'll give them a bit of time to sort this stuff out before I make a video saying they suck. :)
@@ApertureGrille Thats very interesting because i Felt The same way about raw input even tho it goes agaist everything ive learned / experienced in The Past. Raw input have mainly been a tool to get rid of mouse acceleration to me but in this case it Felt better, Probably due to less input delay then.
Bean Human already mentioned that you can create a toggle bind for changing your frame caps in Apex, but you can also change your binds on the fly by creating a key bind that reloads autoexec.cfg mid-game, assuming that you added +exec autoexec to launch options. Example autoexec.cfg: bind "f11" "exec autoexec" bind "f12" "toggle fps_max 180 240 0" This way you can change your frame rates in the autoexec.cfg file and reload it in game by pressing f11 if you need to. I hope this helps.
i've been searching ways to lower input lag while playing csgo and valorant and i had no idea not maxing out your gpu helps. thanks my man i'll test it out, great video!
Thank you very much for this informative video. In battle(non)sense's video, there was no difference in latency between an fps cap of 60fps vs a GPU-limited 60fps with reflex ON. This led him to conclude that as long as a game has reflex, you can pretty much forget about fiddling around with the fps cap. Your tests in this video show that the fps cap is still necessary if you want the absolute lowest latency. Assuming there was nothing wrong with battle(non)sense's measurements, the only reason he didn't see a latency difference is that, unlike you who tested on the lowest graphical settings at 360p and 720p(in Warzone), battle(non)sense tested at 1080p *High* settings with 95% GPU usage. By the way, in apex legends, there is a weird framerate/latency issue above 190fps. Respawn said there were gonna fix it but I don't think they fixed it yet. Also, according to your GPU usage bar chart 9:07, the jump from 60% to 70% to 80% to 90% only increases the latency by 1 or 2 milliseconds. so, that 80% GPU usage at 220fps in apex wasn't the main cause of that latency issue.
If I'm looking at his results table correctly, that seems to be the case. In Fortnite, his 60 FPS cap situation was at 95-96% GPU load, but I'm still surprised there wasn't a greater difference. For my testing, I wanted to have a very clear distinction between the CPU and GPU tests, so that's why I used such low resolution scaling for the CPU results. For Destiny 2, 16:01, at my 60 FPS capped settings, my 1070 Ti was at an average of 21% use. That's obviously a bit unrealistic (hah!), but I was interested in trying to isolate the game engines from the GPU. A few people have pointed out the 190 FPS issue in Apex, and I saw a linked RUclips video of what looked like stutter when the game was past 190 FPS, but I didn't notice any hitching when playing around, and I'd like to think I'm hyper-sensitive to it. Even with V-Sync off, uneven frame pacing is pretty easy to spot. And all my data at 220 FPS falls nicely into a proper uniform distribution. Right now, I've sold off all my monitors, so I can't test 200 FPS while in the VRR range of a 240 Hz monitor, but I should have a new one coming in soon, so I can check that then.
Awesome video. As someone who plays both games, it would be interesting to see the difference in mouse latency in Quake Champions (CPU limited) vs. Quake Live. Also interesting, a comparison of Quake Champions bad internal frame limiter to external limiters.
I love Quake Live, but it's also frustrating. The com_maxfps setting only caps to integer divisors of 1000, so if you try to cap to 60, it'll actually cap to 1000/16 = 62.5 FPS and you can't cap to 120, since it'll cap to 1000/8 = 125 FPS. And perhaps you've noticed this as well: you can't strafe smoothly. If you hold A/D to move left or right, the camera doesn't pan smoothly; it kind of jumps and hitches, and it drives me crazy! But it might be fun to take a look at the two Quakes to see how they compare.
@@ApertureGrille I may be wrong, but if you are referring to the view bobbing, you can turn that off with cg_bob "0" . Similarly for screen shake on damage taken, cg_kickScale "0" . Both of these have a default value of 0.25 afaik. Also, as you likely know, the gun model can be made stationary from the GUI weapon settings.
Any chance you could test some older games like CS:GO and TF2? Curious to see how Source Engine holds up for input lag. Your channel provides amazing data that cuts through the placebo effect surrounding input lag and monitor response times.
I tested CS:GO here: ruclips.net/video/8ZRuFaFZh5M/видео.html It's not exactly the same, since for that video I was mostly looking at how games behaved with adaptive-sync on, but it should give a general idea. CS:GO did fantastic with its internal cap at 165 FPS. It would do even better if you let the framerate run into the 300+ range.
I would really like to see tests of RTSS scanline sync. With that, the game renders without Vsync, but is paced by RTSS in a such way that keeps the tearline at top/bottom of the screen. In my tests 30 hz tests, it performs better than Vsync, but not as well as a good in game frame limiter (in Spellbreak). It gets rid most of the tearing though.
I think the answer is 99.9%... The gpu is just a machine that produces and queues up frames. Theres no technical reason why it should start having problems UNTIL it reaches that overloaded threshold of 100% usage. The second the gpu is overloaded to 100%, it has frame queuing issues because it can’t keep up, and that is what introduces the lag and delay that you ultimately feel. I would take this videos advice to try to aim for 90% usage or less, to stay on the save side. Especially in competitive games. But in reality, 99.9% usage shouldn’t feel any different then 1% usage
At 9:07 I show a chart from my previous video that keeps a static framerate but adjusts the GPU load down in 5% steps from 100%. The biggest decrease is from 100% to 95%, but as ThisIsExile91 says, that happens because I'm not hitting the GPU wall. All the further reductions come from quicker GPU frame times.
1) CS:GO you might as well turn low latency mode to OFF unless you have high GPU usage. Battle Nonsense test's showed that low latency mode ULTRA/ON actually has higher latency than using OFF when you have very low GPU usage. 2) Valorant has reflex + boost so you might as well leave it enabled. As far as I know, Reflex takes precedence over low latency mode.
@@fusedOG there is no reason to use +boowt given that it just diables the internal power limiter of your gpu in a game that has an abundance of frmaes to begin with the potential shortening if your cards lifespan aint worth the minimal boost
This was an another amazing video, thank you so much for the information share. I did had one question - have you tried using external frame rate cap tools like RTSS, or Nvidia's max frame rate? How does it differ for input latency?
Check out my video here: ruclips.net/video/8ZRuFaFZh5M/видео.html Internal caps work best (except for Quake Champions, which I learned for this video), external caps are all similar, but a bit worse, and RTSS's scanline-sync comes in last.
Fantastic video. Keep them coming. For future testing of Apex you could create a toggle bind for the fps_max command from the launch option. For example "+bind x "toggle fps_max 60 100 144"". Also you can lock the adaptive resolution in apex to the lowest value by modifying the videoconfig.txt thats hidden somewhere in the appdata folder. Set "setting.dvs_enable" to 1 to enable adaptive resolution and then "setting.dvs_gpuframetime_min" "setting.dvs_gpuframetime_max" both to 1 or 0, however this does not work without TXAA enabled and is probably not a very reliable variable in testing. It's also possible to create a bind that toggles this function on and off instantly ingame, but I'm not sure that still works, they've been hard at work locking down the settings. In Warzone I'm pretty sure you can set render scale to 33% in the config file, might bring you up to acceptable levels for comparisons. If you're taking suggestions on the topic of latency, since you seem to value low latency. I would really enjoy a comparison of latency between games that have both a Linux and Windows client such as CS:GO, or a comparison between Overwatch on Windows and through WINE/Proton under Linux(Lutris can set this up with no effort required). The reason being that in my experience the latency and smoothness of these two games were WAY better under Linux(Pop!_OS) and I'm in DESPERATE need of a sanity(placebo) check. Theres also a bunch of Low Latency kernel mods and weird(but easy) optimizations to try out if it's a rabbit hole that interests you. My theory is that Windows forces some v-sync behaviour with the Desktop Windows Manager(DWM) which Linux doesn't and WINE doesn't translate, leading to overall lower latency. In my mind it's like the opposite of "optimizing" a windows installation, you just give it the bare minimum it needs to run. Again, your articles and videos are extremely well put together and an amazing resource to have, keep it up.
Great comment. Maybe I'm a dummy, but I can't get the Apex FPS toggle to work. I'm using the Steam version, if that makes a difference. I did play around with the adaptive resolution setting in the AppData folder and set it to Read Only, but at that point, I was so annoyed with Apex that I just went with what I had :) Destiny 2 was a joy to work with in comparison. I don't know much about Linux (I have a Pop! install on one of my old basement PCs, that's about it), but I'm not the type to go down the lag rabbit hole. And it's mostly because of this: www.aperturegrille.com/reviews/DellP1130/Images/Full-Screen-CRT-Response.png That was done using the same PC and same method of testing I used in this video, but this time with my CRT and a DX9 program running at ludicrous FPS. Nothing special about the PC, no Windows or BIOS settings tweaked, no HPET modification, etc, and the average lag was 0.6 milliseconds before some part of the screen showed white. There was even a response at 0.33 ms! That's why I mentioned framerate is king for lag, and I've been less concerned with Windows specific tweaks. The fact that I can have my micro-controller send a keypress and then see the result happen on screen only 0.6 ms later with my stock Windows install made me realize almost all substantial lag has to do with framerate, GPU load, and game engines. But if I can figure out the Linux CS:GO stuff, I'll play around with it. I've had quite a few suggestions for things/games to test, and I need something for another video... But check out FR33THY on RUclips. He's got a custom latency measurement tool, and he seems to thrive on delving into the nitty-gritty.
Great video. Although I am still questioning my set up also. And it’s affecting my mental and game performance. So here is my pc configuration: ● Processor: 10th Generation Intel® Core™ i7-10700 Processor with vPro™ (2.90 GHz, up to 4.80 GHz with Turbo Boost, 8 Cores, 16 Threads, 16 MB Cache) ● Operating System: Windows 10 Home 64 ● Memory: 16 GB DDR4 2933MHz (2 x 8 GB) ● Hard Drive: 512 GB PCIe SSD ● Graphics: NVIDIA® GeForce® RTX™ 2060 6GB ● WiFi Wireless LAN Adapters: 802.11AX (2 x 2) & Bluetooth® 5.0 ● Power Supply: 650W ● Cooling System: 150W Air Cooling Monitor: Lenovo Legion Y25-25 1080p , 1ms, IPS 240hz. Freesync/Gsync compatible I just want fast snappy gameplay like the pros. Any suggestion to make it how they have it? Or close to? I am always tweaking the settings. I just want to do a setting and leave it. I have gsync + vsync in NVCP turned on about to try that. Hz set to 240. And FPS capped at 160 on apex. But I know I’ll end up adjusting again because it’s messing with me!
From my knowledge, dynamic resolution scaling prevents the gpu from running full tilt and keeps it around 90-95%, do you think this could be a solution for preventing high input lag and maxing out of the gpu in games that support it?
Is there any reason to frame cap AND turn on Reflex? Let's say the frame rate met the cap 95% of the time but dipped below on occassion. Would Reflex be "picking up the slack" in those situations and kicking in? Or is it better if we set a cap and just shut off Reflex. I am definitely sticking with in-game capping. My only question is to also have Reflex on or off.
If a game has Reflex, I'd always turn it on. It never hurts. You're right about the "picking up the slack" part. Use an in-game cap to keep your GPU utilization under control, but for those 5% of times when the GPU use does hit 100%, Reflex will really help cut down on latency.
Before I've finished watching, I think your test-tool is great, but with such a trivial scene, the required CPU is sub 1ms, in a modern game, CPU bottleneck make input latency bad because they have to sample the mouse and keyboard a good 10-30ms before they update the graphical display, especially in a game with a native 30hz tick. I wouldn't be surprised if some games, with less competency, would have a bad framerate limiter causing delays like vsync too.
Have you tested between scenarios like [vrr + vsync @ 117fps & 120hz] vs [vrr @ 120fps & 240hz] ? Also, Diabotical has a multithreaded input where I'd like to see latency differences. Excellent test setup and results btw.
I haven't specifically tested those, but for VRR, [119 FPS on a 120 Hz monitor] vs [120 FPS on a 240 Hz monitor] will have the same latency at the top of the screen (if both monitors had equal lag), but the 240 Hz monitor paints its image from top to bottom in 4.16 ms vs the 120 Hz monitor, which takes 8.3 ms to scan the image from top to bottom, so in effect, the 240 Hz monitor wins. For VRR displays, having a really high max refresh rate allows for something akin to Quick Frame Transport, which is really just how fast the image is painted from top to bottom. I downloaded Diabotical, but I haven't played much of it. It'll take me a bit to get used to the movement. I'm still used to Q2/Quake Live physics.
How would you test USB-peripheral-to-first-pixel-response (system end-to-end) latency if you're using 4-way SLI, which doesn't have the option to use ULLM in NVIDIA driver control panel? Also, in my case, I'm using SFR or CFR SLI rendering modes, not AFR, AFR of SFR, or (FS)AA SLI rendering modes. Latency does not increase on multi-GPU setups when utilizing SFR/CFR rendering modes, latency remains identical to single-GPU AFAIK. Also, what is the difference between using in-game FPS limiter, versus the FPS limiter in the NVIDIA driver settings? I always use that preferably to VSYNC to reduce judder and screen tearing. P.S. In Crysis 3, with maximum settings including 8X MSAA at 1920x1200 resolution, using my 4-way GTX 1080 Ti SLI, 4 GHz Core i7-6950X setup, I can get over 240 fps with the FPS limiter off and I don't use VSYNC. In some areas in GTA IV with max settings, I could get 300-600+ FPS in 7680x4800 resolution, in Skyrim: Special Edition (vanilla, no mods) with maximum settings I got 80 FPS average at 7680x4800 resolution, in Crysis Remastered I was able to get 45 FPS average with maximum settings (Can it Run Crysis? mode, including raytracing) at 3840x2400 resolution. Shadow of the Tomb Raider, over 70 FPS average with maximum settings at 3840x2400 resolution (DXR for shadows was turned off).
I'm sad I found your channel months after you stopped uploading - hopefully it's a temporary break. I would love a video on your response time setup, explaining all the hardware and software so that we can do our own measurements at home.
I'm still here! I just don't have anything to review at the moment. But if you want to see more detail about how I'm testing, check out my 10% to 90% GtG video, where I cover response time testing. In that video, I go over exactly what I use to measure the screen. It's really just a micro-controller and a light-to-voltage probe, but the set up is a little bit complicated.
Would it be possible to revisit Apex Legends purely for Nvidia Reflex? They now added an in-game option to either disable reflex, enable reflex or enable reflex with boost. It would be very interesting to see if Reflex improves the game performance and response time or if you are better off leaving it off or on boost or normal reflex.
Reflex wins because it allow the game to run at higher FPS as long as the GPU utilisation < 100% whereas ULLM caps the FPS just below the monitor's refresh rate. Not to mention if both are enabled, Reflex overrides ULLM according to Nvidia
Can you answer me should I enable latency mode in cvar for destiny2 ? if yes then 1 or 2 should I also limit fps to 200 or 240 in 240 hz monitor ? Thanks
I heard Quake Champions might hve some kind of anticheat that cause issues with external limiters (atleast if they have overlay). this was just after they released Reflex.
great vids! curious if you tried 144hz with elmb sync on v sync off with the asus vg27aq? it doesn't flicker at 144hz but does when set to 165hz... i have the asus vg27bq is a tn panel but faster response time 0.5ms... any thoughts if these monitors perform the same? thanks!
I have an unrelated question and not sure if you know the answer: does UE4 have an internal engine limit of around 1000 fps? I tested it just with a test scene (self emissive cube, no light) in forward rendered mode and it seems to bounce off around 1000 fps or 1ms frame time. In Unity I can get around 2500fps with a similar test scene. Further in UE4 my GPU doesn't seem to even break a sweat whereas in unity at 2500 the thing spins up like a mofo. Could just be CPU bottlenecked. I'm now on a mission to write custom openGL, vulkan, and DX 11 and 12 test scenes to see if I can find what can run the fastest. Using an RTX 3090 I managed to get for a wallet destroying sum of money lol.
I've been using the internal cap for 1000 FPS, but uncapped, it hovers around 1100-1200 FPS on my 7700k at 4.7 GHz. I don't think Epic was really thinking about optimizing for 1ms frame times, so there's a lot going on in the background, even on simple scenes. If you're searching for something that runs at 5000+ FPS, check out github.com/NotThat/Benchmark-DirectX9/releases Grab DX9.Black.White.exe. The guy who did these, NotThat, was doing something similar as well, writing DX11 and DX12 "games" to see which was the fastest.
If you're interested, could you investigate how in a game like csgo, blurbusters and some 240hz monitor owners recommend you turn on gysnc, and turn on v-sync in the nvidia control panel. Then turn off the v-sync ingame, and capping the fps 3-5fps lower than how much hertz the monitor has. This forces the game to always utilize g-sync, and won't have the stutter of the v-sync when the fps crosses the 240fps threshold, thus there is no additional input lag while using gsync.
I cover a bit of that in this video: ruclips.net/video/8ZRuFaFZh5M/видео.html But I'm not quite sure why people recommend setting V-Sync with the nvidia control panel rather than using the game to set V-Sync. For instance, in UE4, using in-game v-sync and its internal fps cap works great. At 4:49 in this video, I show UE4 running that situation. V-Sync was set to ON for those measurements. The only reason the lag is slightly higher for those measurements than with v-sync off is that we aren't getting partial, torn frames. Otherwise, the internal v-sync setting has no impact on lag. I'll see if I can take a look at why people are recommending using the nvidia control panel v-sync option. I've heard that before as well, and I can't think of why it would make lag better.
@@ApertureGrille I think it was more on the lines of "the correct way to use g-sync". People still argue over what's best for a game like csgo, purist say that the best is still pure uncapped no syncs whatsoever to push as much frames for absolutely lowest input lag. But then, those that recommend using g-sync argue that input lag is either neglible or unaffected and the benefits or reduced motion blur and/or a tear free experience outweigh the cost of capping the fps. I've seen it done for other games like apex and overwatch. They say people could still argue the former at 140fps at 144hz, however it gets more interesting at 236fps at 240hz, I'm even curious to see how an overclocked panel like the asus can handle it at like 276fps at 280hz.but yeah, most of it are just claims of this and that, no one really has good numbers for comparison or the right implements to make a proper conclusion. But from what i can gather tho, at 360hz, i think there would be no argument for keeping g-sync on, and input lag would still be at comfortable competitive levels.
How you get 220 fps in Apex with 1070ti. I have the same card and I struggle to have stable 120. When I'm fly in the ship over the map I get below 100 fps. And I'm already running at low and even went to 1080p on my 1440p g-sync monitor.... what do I do wrong?
Hi a5hun, love your detailed analysis. Btw, have you ever noticed a horizontal line flicker for a millisecond (I'm talking so quick that you question whether you're imagining it or not) on your Legion y27q-20? I've had mine for a little over a month and at first I thought I was seeing things but lately I'm starting to think it's an actual issue with the display. I tried disabling gysync and lowering hz to 144.
Nothing like that. I've got it sitting on my desk right now, and I've been using it since Lenovo sent the review unit in December. That's a hard thing to diagnose; there was a point last year where NVIDIA's drivers were causing flickering artifacts like that; I was seeing it happen on my VG27AQ, but they eventually fixed it. So it could be the driver or GPU memory or the DisplayPort cable.
@@ApertureGrille Realized I'm using an old driver. Will see if updating will fix it. I also have the vg27aq as a secondary monitor and it doesn't have this issue.
I haven't tested the LG 27GL850, but from other reviews, it should perform about the same as the Lenovo Y27q-20 and Dell S2721DGF. They all use the same nano-IPS panel, so you're mostly just paying for different industrial design and slight implementation differences. I think the Lenovo and the LG both have working sRGB emulation modes, which the Dell lacks, but you might just want to get whichever you can find at a decent price.
Can you test Overwatch 2 ? I wonder if i should let fps go to max with 100% utilization and enabled reflex boost or should i cap at 116 fps, gsync, low gpu load.
Try this: ruclips.net/video/8ZRuFaFZh5M/видео.html At around 7:41, I've got the full chart with NVIDIA's FPS limiter, LLM, ULLM, and RTSS. All the external limiters perform a bit worse than UE4's internal limiter.
ayo can you review a setup i wanna buy some day and can you spot out any flaws? i could reply with the setup or anything i should buy instead of one thing?
I wonder why you tried the Steam version of Apex. Doesn't the Origin version allow you to toggle NV Reflex on/off with a launch option or did I miss something?
@@ApertureGrille I could be wrong though, don't have a way to support my claims. I do know that I'll be downloading Apex on Steam, thanks for making this video!
Hi. What is the best setting please for warzone: G sync: on V sync: fast Latency mode: ultra Cap frame rate to: 350 fps I have 3090 GPU, Intel I9 11900k CPU and 360hz asus monitor GSync Please give me the best setting. Thanks so much 🙏🏽
Yes! At least if the game has implemented it properly. There's a chart here: ruclips.net/video/8ZRuFaFZh5M/видео.html That shows a comparison of UE4's internal limiter, which is excellent, versus NVIDIA's control panel limits, ULLM, and RTSS's limiter. Internal does best, and the external caps are all roughly the same.
I just found your channel after getting a 360hz monitor, do I need to worry about 100% gpu usage at these frame rates and is there an easy way to limit my gpu usage to say 90% as I am sure I am maxing mine out trying to get 360fps. My monitor has Gsync
I think the Destiny 2 chart at 16:10 illustrates the high refresh rate behavior the best. At 220 FPS, running at 100% GPU use only increase the latency from 14.1 ms to 18.8. That's not so bad, and the difference, I think, would be even smaller at 360 FPS. Would anyone notice that 4.7 ms? Not sure. With modern games, your framerate will naturally fluctuate based on the scene complexity, but I think it's a good idea anyway to set an internal framerate cap to something that'll keep you away from that 100% case most of the time. Even if you have to cap to 320 FPS, that's only a frametime difference of 0.347 ms compared to 360 FPS, whereas running at 100% use at 360 FPS might increase that by five times. I know it's a bit irritating that you can't use all the GPU you've paid for without getting a lag penalty, but that's what we've got. Internal caps work best. I'd also recommend using G-Sync. The lag penalty over V-Sync off is not that large, but you get a massive benefit with the reduction of screen tearing and judder. VRR is amazing, and I always recommend people use it... unless you're playing CS:GO at 700 FPS. :)
@@ApertureGrille I thought you might like to know my results (although I am about as novice as novice gets), with my gpu at about 60% usage and my pc running between 290 and 340 fps in siege with gsync on, I am able to use the nvidia reflex monitor built into my monitor and my pc+display latency results are normally between 10-12ms.
@a5hun quick question. Does this mean ullm does bad things when you run an old game at it's max capped framerate? I'm thinking about QuakeLive which doesn't use 100% of the GPU and CPU. For csgo I use ullm with my 6600k at 4.5ghz and 1080ti (1080p smaa 8x and aniso 16x shaders on low and textures on low) on some new maps I hit 90 fps and I really feel ullm helps reducing input lag.
ULLM only works on DX11 titles, so unfortunately it won't do anything for OpenGL/Vulkan/DX12 titles, and that includes Quake Live. I haven't actually tested Quake Live for lag, but QL is a weird one because it's not as adaptable as other games; it tops out at 250 FPS, and the framecap only caps to integer divisors of 1000 (so 1000 / 8 = 125). For QL, I'd just turn V-Sync off and run at 250 FPS. You'll get tearing each frame, but there's not much that can be done. How are you getting such low framerates in CSGO? Your PC is pretty great. I don't think ULLM will hurt in CPU bound scenarios, maybe 1 ms or so, but it's only really useful if you're 100% GPU bound.
@@ApertureGrille yeah I have no idea how I am getting so bad framerate in CSGO. Is there a benchmark that I can use to test within this game? As for QL, I use a reshade motion blur shader that completely remove screen tearing at the cost of 0.4ms GPU time. It feels great.
But how much of that does actually matter, when human reaction and other factors like internet latency and the consistency thereof is probably more wonky than 3ms difference (3ms taken from the QC graph as difference between 420fps max and 240fps cpu bound). AFAIK human reaction, depending on what happens i.e. expected/prepared events vs. unexpected/unprepared events and mental stack, the variance is probably WAY bigger than that. Second thing is how certain interactions are animated i.e. how many frames and which action are possible (I'm coming from a fighting game, if that isn't already clear ;-) ) Like the time a shotgun or a rifle is reloaded is going to be the same in seconds, but no in frames, obviously. While I certainly agree to never play GPU bound at 60fps and rather 120 or even 240 if possible. Further bigger improvements, especially in the beginning, are easier done in the game i.e. getting better, decision making, etc. than adjusting anything else outside the matches boundaries. Otherwise there really isn't anything to take seriously, if the gap is going to be that big with highly different machines. BUT, that wasn't the point here. I know.
Good points. Chasing down a 3 or 4 millisecond reduction in latency isn't particularly worth it if it costs $500 or you have to reduce the rendering resolution so low that you lose visibility. Check out this program flood posted on the Blur Busters forums that lets you A/B test for lag: forums.blurbusters.com/viewtopic.php?f=10&t=1134 I think most people would have trouble correctly determining any latency below about 20 ms.
Because I was testing with a QHD monitor, even with the resolution slider at 50%, my 1070 Ti isn't fast enough to prevent that 99% use when measuring at 220 FPS, so I dropped the CPU-limited (or capped) results. It's not ideal that I'm testing with a 1070 Ti, and I was hoping to get something more powerful, but video card prices are out of control right now. A few people mentioned that you can adjust that rendering resolution lower by editing the config files, and I may try that for the next video. And Apex Legends recently got an update with a working Reflex ON/OFF toggle, so I can test that as well.
@@ApertureGrille No my point was not resolution slider. You should use framerate limeter inside warzone. For example if your gpu can render 220 fps with 99 % load just set fps inside warzone to 180 fps and load on your gpu decreases. Look at the video the fps cap is unlimeted.
@@alikalhor77 Oh, I understand! For the testing, I was trying to hit three consistent framerate targets amongst all the games: 60 FPS, 120 FPS, and ~220 FPS. That way I could make a fair comparison at the end. For all the other games I tested, I was able to reach ~220 FPS without running into a GPU bottleneck, but with Warzone, that same 220 FPS was hitting the GPU limit. So instead of capping at something lower like 180 FPS, I decided not to include the CPU limited tests; that would have had me comparing 60-120-220 for every other game with 60-120-180 for Warzone. 180 FPS vs 220 FPS is only around a 1 ms frame time difference, so it wouldn't have been that unfair, but I think people might have gotten mad at me if I compared the two when they're 40 FPS different :)
@@ApertureGrille very good content all around. the video from battlenonsense suggested that reflex works exactly the same as not gpu bound scenario in terms of latency but your findings shows difference between the two . I think it's about implementing and and result will be game dependent. He tested overwatch and fortnite I think 🤔
@@alikalhor77 I think that difference may be down to how absurdly low I dropped the GPU load. For instance, I don't think anybody is going to be playing Destiny 2 at 25% resolution scaling. At that setting, my GPU was only around 38% load. So while that's not a realistic use case, I think it does help illustrate how quickly the game engine can react if it's not being limited (or barely limited) by the GPU. It's kind of like doing CPU tests at 720p.
Awesome video though I'm still questioning my settings. I have a gtx1650 4gb on ryzen 7 2700x .still don't know if I should have reflex off or on or on+boost on warzone
For games that have reflex, there's no downside to turning it on. Reflex reduces lag when you're GPU bound, but doesn't hurt when not. But even still, you should try not to hit 100% GPU use. With your 1650, that might be hard because Warzone is pretty heavy on GPU, but try to turn down settings and cap the framerate to something you can reasonably maintain.
I have always wondered if relex increases the amount of time it takes to see something on the screen for you to respond to. It seems really hard to test that though. Like for example with reflex off would you see that guy coming around the corner sooner than you would with it on? Maybe not, but I don't know enough to say really.
Low-ish framerates and high GPU use! For DX11 games, it does "OK" at 120 FPS. In Destiny 2, ULLM cut the lag from 41.2 to 35.9 ms in a GPU limited scenario; that's not a ton, but it certainly helps. But you're right, it does best at ~60 FPS. Generally, ULLM is a good idea if the game you're playing doesn't have a good internal cap, so you keep bumping into that 100% GPU wall.
@@ApertureGrille Ah okay that makes sense. Are there any cons to using it? I'm a QuakeC main and get a pretty consistent 200+ fps (using Rivatuner to cap at 280fps) thanks!
@@FabFeline I should have tested QC, but for Destiny 2 at 200+ FPS, it looks like it may hurt slightly... maybe less than 1 ms. Not something I'd worry about.
When the video went up, this NVIDIA page: www.nvidia.com/en-us/geforce/news/nvidia-reflex-apex-legends-out-now/ said that the +gfx_nvnUseLowLatency cvar didn't work due to a bug. The season 8 patch came out only a few days after I posted the video, and they finally added Reflex in the menus! Haha! Just in time, Respawn! :) I need to hop back in and re-test it now that it can be turned off. And people were also saying that Apex behaves badly above 190 FPS, so I want to check that out as well.
@@ApertureGrille People may be right, I can play with 160-190 fps, but there is some stuttering on my system in 1440p (10700k, rtx 2070, 3800mhz ram, nvme ssd). And sometimes I get huge fps drops to 40-50 fps in some scenes. If fps is stable 140 the input lag feels inconsistent.
@@ApertureGrille Please, if you will do comparison in R6 Siege, make comparison of versions DX11 with fps limiter and ULL vs Vulkan + Reflex and limiter. For my feelings DX11 version seems to be more responsible than Vulkan with reflex, which was surprise for me.
the advice here is weird cause...if you disable vsync in most games, your GPU will run at 100%.....but at the same time you're saying NOT to run your GPU at 100%........so I guess set a frame cap that's over 200fps but not so high that you'll max out your GPU usage...?
That would work, but there are other ways as well. For something like Quake Champions, if you drop the render resolution a bit (not to the crazy 25% I was using), in real gameplay like a multiplayer match, you'll hit CPU limits instead of GPU limits. If I drop my rendering resolution down to 80%, in bot matches, I'm usually hovering around 150 FPS because of CPU limits, and the GPU is sitting at 85 to 90% use.
what is point in measuring new frame which is only at bottom, playing game it is useless information, what is middle is what you aim, so only when this part of screen is refresh it gives you useful information.
Obviously very late but HOW in the fuckin world is your gpu running at 35 C at 70-80 usage when running that mess of a game named "Quake Champions"?? You got your pc outside in Alaska or what homie?
Pretty hefty undervolt on the GPU (0.881V), but I also have the Strix 1070 Ti running with liquid metal. The liquid metal makes a huge difference, so much so that any GPUs I have that have a nickel-plated heat spreader are getting the LM treatment. The 7700k also has LM under the heat spreader.
you havnt even scratched the surface . QC has very strange input lag that effect everything mouse. movement. desync of opponent model. the worst shooter from technical side i have ever seen
criminally underrated testing
Fantastic video! Can't believe I've just come across it. Will link it in my upcoming video.
You are answering all the questions I’ve been trying to answer for years!’
Awesome stuff. I think lag analysis videos on a per game basis would work great - at least I would love to see your indepth take on games like Rainbow Six Siege and Valorant. I'd also love to see RTSS included.
found your channel few days ago and im binging every single video they are just so well made there is no one who does this type of stuff better than you. hope some day you get the recognition u deserve
Thank you! The first few from 2019 were a little rough, but I'm learning.
Keep the great work and create a Twitter man ! Love to follow you
Wow, great video! Love your stuff.
Amount of work and time required to produce this video is insane. I hope you will get more attention in a short time.
There is a bug in apex that halves frames above 190fps, which would explain your slightly higher readings. Thanks for the video!
wait what for real? so should i cap it?
@@noel4140 Yep! Cap to 190, it should slightly reduce input latency and make the image a whole lot more stable. Try it yourself using RTSS, it should be very apparent, particularly if you go back and forth from 190 to 240 or higher. Theres a reason why most pros keep a 190 cap.
Amazing work. I literally was thinking about this for the last decade and just didn't fully understand before this!
Thank you!
Another detailed and informative video. Good job.
So, in terms of input lag, the best combination for players who don't use blur reduction is:
- Highest possible framerate
- GPU utilization under 100%
- Nvidia Reflex (or ULLM)
And for players using blur reduction (backlight strobing, CRT monitors):
- Framerate always above refreshrate
- RTSS Scanline Sync with tearline near the bottom of the screen
- Fast Vsync on
Scanline Sync is fantastic for CRTs. I wish more people could see a proper, glass-smooth game on a CRT.
FYI - Quake Champions PTS supports Raw Input and Nvidia Reflex + Boost now. It'd be pretty cool to see another input lag analysis when the patch hits live servers.
Thanks for the tip! I just re-ran all the same measurements I did for Destiny 2 on QC PTS; I was planning on doing another video specifically about Quake Champions, but... Reflex in QC is literally doing nothing right now. On, Off, On + Boost... makes no difference. In a GPU limited scenario at 60 FPS, all three of those modes return the same lag of around 57 ms. I'm willing to cut them a little slack since they haven't introduced Reflex into the main game yet, but it's a bit annoying that they have an option that doesn't work. Although, I do remember there being a period of... maybe months where the V-Sync option in the settings menu also didn't work.
Interesting, though, using raw input for the mouse makes lag worse. Setting the mouse back to DirectX input dropped the lag from 57 ms to 44 ms.
I'll give them a bit of time to sort this stuff out before I make a video saying they suck. :)
@@ApertureGrille Thats very interesting because i Felt The same way about raw input even tho it goes agaist everything ive learned / experienced in The Past. Raw input have mainly been a tool to get rid of mouse acceleration to me but in this case it Felt better, Probably due to less input delay then.
Thank you for the destiny 2 measurement. Almost nobody has done this as far as I know.
Bean Human already mentioned that you can create a toggle bind for changing your frame caps in Apex, but you can also change your binds on the fly by creating a key bind that reloads autoexec.cfg mid-game, assuming that you added +exec autoexec to launch options. Example autoexec.cfg:
bind "f11" "exec autoexec"
bind "f12" "toggle fps_max 180 240 0"
This way you can change your frame rates in the autoexec.cfg file and reload it in game by pressing f11 if you need to. I hope this helps.
Thanks for new amazing video , keep up with great work
extremely underrated channel
i've been searching ways to lower input lag while playing csgo and valorant and i had no idea not maxing out your gpu helps. thanks my man i'll test it out, great video!
As a reviewer myself I highly appreciate your content, it helps a lot!
Thank you very much for this informative video. In battle(non)sense's video, there was no difference in latency between an fps cap of 60fps vs a GPU-limited 60fps with reflex ON. This led him to conclude that as long as a game has reflex, you can pretty much forget about fiddling around with the fps cap. Your tests in this video show that the fps cap is still necessary if you want the absolute lowest latency. Assuming there was nothing wrong with battle(non)sense's measurements, the only reason he didn't see a latency difference is that, unlike you who tested on the lowest graphical settings at 360p and 720p(in Warzone), battle(non)sense tested at 1080p *High* settings with 95% GPU usage.
By the way, in apex legends, there is a weird framerate/latency issue above 190fps. Respawn said there were gonna fix it but I don't think they fixed it yet. Also, according to your GPU usage bar chart 9:07, the jump from 60% to 70% to 80% to 90% only increases the latency by 1 or 2 milliseconds. so, that 80% GPU usage at 220fps in apex wasn't the main cause of that latency issue.
If I'm looking at his results table correctly, that seems to be the case. In Fortnite, his 60 FPS cap situation was at 95-96% GPU load, but I'm still surprised there wasn't a greater difference. For my testing, I wanted to have a very clear distinction between the CPU and GPU tests, so that's why I used such low resolution scaling for the CPU results. For Destiny 2, 16:01, at my 60 FPS capped settings, my 1070 Ti was at an average of 21% use. That's obviously a bit unrealistic (hah!), but I was interested in trying to isolate the game engines from the GPU.
A few people have pointed out the 190 FPS issue in Apex, and I saw a linked RUclips video of what looked like stutter when the game was past 190 FPS, but I didn't notice any hitching when playing around, and I'd like to think I'm hyper-sensitive to it. Even with V-Sync off, uneven frame pacing is pretty easy to spot. And all my data at 220 FPS falls nicely into a proper uniform distribution.
Right now, I've sold off all my monitors, so I can't test 200 FPS while in the VRR range of a 240 Hz monitor, but I should have a new one coming in soon, so I can check that then.
Awesome video. As someone who plays both games, it would be interesting to see the difference in mouse latency in Quake Champions (CPU limited) vs. Quake Live. Also interesting, a comparison of Quake Champions bad internal frame limiter to external limiters.
I love Quake Live, but it's also frustrating. The com_maxfps setting only caps to integer divisors of 1000, so if you try to cap to 60, it'll actually cap to 1000/16 = 62.5 FPS and you can't cap to 120, since it'll cap to 1000/8 = 125 FPS.
And perhaps you've noticed this as well: you can't strafe smoothly. If you hold A/D to move left or right, the camera doesn't pan smoothly; it kind of jumps and hitches, and it drives me crazy!
But it might be fun to take a look at the two Quakes to see how they compare.
@@ApertureGrille I may be wrong, but if you are referring to the view bobbing, you can turn that off with cg_bob "0" . Similarly for screen shake on damage taken, cg_kickScale "0" . Both of these have a default value of 0.25 afaik. Also, as you likely know, the gun model can be made stationary from the GUI weapon settings.
Any chance you could test some older games like CS:GO and TF2? Curious to see how Source Engine holds up for input lag. Your channel provides amazing data that cuts through the placebo effect surrounding input lag and monitor response times.
I tested CS:GO here: ruclips.net/video/8ZRuFaFZh5M/видео.html
It's not exactly the same, since for that video I was mostly looking at how games behaved with adaptive-sync on, but it should give a general idea. CS:GO did fantastic with its internal cap at 165 FPS. It would do even better if you let the framerate run into the 300+ range.
This video is amazing! Good job man
keep making content! I love your videos!
Thank you for this informative video. Keep up with the good work)
Great video! I'm curious how this would look like on a 360 or even 390 Hz monitor in CSGO.
I would really like to see tests of RTSS scanline sync. With that, the game renders without Vsync, but is paced by RTSS in a such way that keeps the tearline at top/bottom of the screen. In my tests 30 hz tests, it performs better than Vsync, but not as well as a good in game frame limiter (in Spellbreak). It gets rid most of the tearing though.
I admire the effort that goes into your videos, awesome work. :)
Thank you! Quake Champions has implemented Reflex now on the PTS, so I may re-run those measurements once they release it in the main game.
@a5hun at what percentage of GPU utilization does it start to get worse in terms of input lag? 70+% ? 80+% ? 90+% ?
I think the answer is 99.9%...
The gpu is just a machine that produces and queues up frames. Theres no technical reason why it should start having problems UNTIL it reaches that overloaded threshold of 100% usage.
The second the gpu is overloaded to 100%, it has frame queuing issues because it can’t keep up, and that is what introduces the lag and delay that you ultimately feel. I would take this videos advice to try to aim for 90% usage or less, to stay on the save side. Especially in competitive games. But in reality, 99.9% usage shouldn’t feel any different then 1% usage
At 9:07 I show a chart from my previous video that keeps a static framerate but adjusts the GPU load down in 5% steps from 100%. The biggest decrease is from 100% to 95%, but as ThisIsExile91 says, that happens because I'm not hitting the GPU wall. All the further reductions come from quicker GPU frame times.
So for games like cs go and Valorant that are constantly 300+fps would you still use nvidia ulm AND reflex boost?
1) CS:GO you might as well turn low latency mode to OFF unless you have high GPU usage. Battle Nonsense test's showed that low latency mode ULTRA/ON actually has higher latency than using OFF when you have very low GPU usage. 2) Valorant has reflex + boost so you might as well leave it enabled. As far as I know, Reflex takes precedence over low latency mode.
@@fusedOG there is no reason to use +boowt given that it just diables the internal power limiter of your gpu in a game that has an abundance of frmaes to begin with the potential shortening if your cards lifespan aint worth the minimal boost
@@mm-rw3hi minimal yes but if you're at a pro level then every advantage should matter
@@fusedOG his tests showed that ultra can increase latency at low gpu usage, not "on". He's never tested "on" which was a lot of people's complaint.
This dude is awesome
This was an another amazing video, thank you so much for the information share. I did had one question - have you tried using external frame rate cap tools like RTSS, or Nvidia's max frame rate? How does it differ for input latency?
Check out my video here: ruclips.net/video/8ZRuFaFZh5M/видео.html
Internal caps work best (except for Quake Champions, which I learned for this video), external caps are all similar, but a bit worse, and RTSS's scanline-sync comes in last.
Fantastic video. Keep them coming.
For future testing of Apex you could create a toggle bind for the fps_max command from the launch option. For example "+bind x "toggle fps_max 60 100 144"". Also you can lock the adaptive resolution in apex to the lowest value by modifying the videoconfig.txt thats hidden somewhere in the appdata folder. Set "setting.dvs_enable" to 1 to enable adaptive resolution and then "setting.dvs_gpuframetime_min" "setting.dvs_gpuframetime_max" both to 1 or 0, however this does not work without TXAA enabled and is probably not a very reliable variable in testing. It's also possible to create a bind that toggles this function on and off instantly ingame, but I'm not sure that still works, they've been hard at work locking down the settings.
In Warzone I'm pretty sure you can set render scale to 33% in the config file, might bring you up to acceptable levels for comparisons.
If you're taking suggestions on the topic of latency, since you seem to value low latency. I would really enjoy a comparison of latency between games that have both a Linux and Windows client such as CS:GO, or a comparison between Overwatch on Windows and through WINE/Proton under Linux(Lutris can set this up with no effort required).
The reason being that in my experience the latency and smoothness of these two games were WAY better under Linux(Pop!_OS) and I'm in DESPERATE need of a sanity(placebo) check.
Theres also a bunch of Low Latency kernel mods and weird(but easy) optimizations to try out if it's a rabbit hole that interests you. My theory is that Windows forces some v-sync behaviour with the Desktop Windows Manager(DWM) which Linux doesn't and WINE doesn't translate, leading to overall lower latency. In my mind it's like the opposite of "optimizing" a windows installation, you just give it the bare minimum it needs to run.
Again, your articles and videos are extremely well put together and an amazing resource to have, keep it up.
Great comment. Maybe I'm a dummy, but I can't get the Apex FPS toggle to work. I'm using the Steam version, if that makes a difference. I did play around with the adaptive resolution setting in the AppData folder and set it to Read Only, but at that point, I was so annoyed with Apex that I just went with what I had :)
Destiny 2 was a joy to work with in comparison.
I don't know much about Linux (I have a Pop! install on one of my old basement PCs, that's about it), but I'm not the type to go down the lag rabbit hole. And it's mostly because of this: www.aperturegrille.com/reviews/DellP1130/Images/Full-Screen-CRT-Response.png
That was done using the same PC and same method of testing I used in this video, but this time with my CRT and a DX9 program running at ludicrous FPS. Nothing special about the PC, no Windows or BIOS settings tweaked, no HPET modification, etc, and the average lag was 0.6 milliseconds before some part of the screen showed white. There was even a response at 0.33 ms! That's why I mentioned framerate is king for lag, and I've been less concerned with Windows specific tweaks. The fact that I can have my micro-controller send a keypress and then see the result happen on screen only 0.6 ms later with my stock Windows install made me realize almost all substantial lag has to do with framerate, GPU load, and game engines.
But if I can figure out the Linux CS:GO stuff, I'll play around with it. I've had quite a few suggestions for things/games to test, and I need something for another video...
But check out FR33THY on RUclips. He's got a custom latency measurement tool, and he seems to thrive on delving into the nitty-gritty.
Thanks for another great video! I wonder what amount of lag Windows has, when you're not running a game but simply using desktop. Have you tested it?
really informative! thank you!
Great video. Although I am still questioning my set up also. And it’s affecting my mental and game performance.
So here is my pc configuration:
● Processor: 10th Generation Intel® Core™ i7-10700 Processor with vPro™ (2.90 GHz, up to 4.80 GHz with Turbo Boost, 8 Cores, 16 Threads, 16 MB Cache)
● Operating System: Windows 10 Home 64
● Memory: 16 GB DDR4 2933MHz (2 x 8 GB)
● Hard Drive: 512 GB PCIe SSD
● Graphics: NVIDIA® GeForce® RTX™ 2060 6GB
● WiFi Wireless LAN Adapters: 802.11AX (2 x 2) & Bluetooth® 5.0
● Power Supply: 650W
● Cooling System: 150W Air Cooling
Monitor: Lenovo Legion Y25-25 1080p , 1ms, IPS 240hz. Freesync/Gsync compatible
I just want fast snappy gameplay like the pros. Any suggestion to make it how they have it? Or close to? I am always tweaking the settings. I just want to do a setting and leave it. I have gsync + vsync in NVCP turned on about to try that. Hz set to 240. And FPS capped at 160 on apex. But I know I’ll end up adjusting again because it’s messing with me!
Just found this channel. You know your stuff.
Could you please do a review of the lg 24gn600 monitor? It boasts hdr at 200 euros which is the cheapest one that has hdr i think
damn...welcome back man. been waiting 4 ur monitor since 10 month ago. now knwing u start release video again from HUB .
It is a really good explanatory video! But what about internal frame rate limiter of warzone, is it any good?
From my knowledge, dynamic resolution scaling prevents the gpu from running full tilt and keeps it around 90-95%, do you think this could be a solution for preventing high input lag and maxing out of the gpu in games that support it?
Is there any reason to frame cap AND turn on Reflex? Let's say the frame rate met the cap 95% of the time but dipped below on occassion. Would Reflex be "picking up the slack" in those situations and kicking in? Or is it better if we set a cap and just shut off Reflex. I am definitely sticking with in-game capping. My only question is to also have Reflex on or off.
If a game has Reflex, I'd always turn it on. It never hurts. You're right about the "picking up the slack" part. Use an in-game cap to keep your GPU utilization under control, but for those 5% of times when the GPU use does hit 100%, Reflex will really help cut down on latency.
Before I've finished watching, I think your test-tool is great, but with such a trivial scene, the required CPU is sub 1ms, in a modern game, CPU bottleneck make input latency bad because they have to sample the mouse and keyboard a good 10-30ms before they update the graphical display, especially in a game with a native 30hz tick.
I wouldn't be surprised if some games, with less competency, would have a bad framerate limiter causing delays like vsync too.
This is so awesome!
Have you tested between scenarios like [vrr + vsync @ 117fps & 120hz] vs [vrr @ 120fps & 240hz] ?
Also, Diabotical has a multithreaded input where I'd like to see latency differences.
Excellent test setup and results btw.
I haven't specifically tested those, but for VRR, [119 FPS on a 120 Hz monitor] vs [120 FPS on a 240 Hz monitor] will have the same latency at the top of the screen (if both monitors had equal lag), but the 240 Hz monitor paints its image from top to bottom in 4.16 ms vs the 120 Hz monitor, which takes 8.3 ms to scan the image from top to bottom, so in effect, the 240 Hz monitor wins. For VRR displays, having a really high max refresh rate allows for something akin to Quick Frame Transport, which is really just how fast the image is painted from top to bottom.
I downloaded Diabotical, but I haven't played much of it. It'll take me a bit to get used to the movement. I'm still used to Q2/Quake Live physics.
No wonder why QC ever feels play able to me. I just cant Coop with input lag, many decads of low input game play is hard to relearn.
How would you test USB-peripheral-to-first-pixel-response (system end-to-end) latency if you're using 4-way SLI, which doesn't have the option to use ULLM in NVIDIA driver control panel? Also, in my case, I'm using SFR or CFR SLI rendering modes, not AFR, AFR of SFR, or (FS)AA SLI rendering modes. Latency does not increase on multi-GPU setups when utilizing SFR/CFR rendering modes, latency remains identical to single-GPU AFAIK.
Also, what is the difference between using in-game FPS limiter, versus the FPS limiter in the NVIDIA driver settings? I always use that preferably to VSYNC to reduce judder and screen tearing.
P.S. In Crysis 3, with maximum settings including 8X MSAA at 1920x1200 resolution, using my 4-way GTX 1080 Ti SLI, 4 GHz Core i7-6950X setup, I can get over 240 fps with the FPS limiter off and I don't use VSYNC. In some areas in GTA IV with max settings, I could get 300-600+ FPS in 7680x4800 resolution, in Skyrim: Special Edition (vanilla, no mods) with maximum settings I got 80 FPS average at 7680x4800 resolution, in Crysis Remastered I was able to get 45 FPS average with maximum settings (Can it Run Crysis? mode, including raytracing) at 3840x2400 resolution. Shadow of the Tomb Raider, over 70 FPS average with maximum settings at 3840x2400 resolution (DXR for shadows was turned off).
I'm sad I found your channel months after you stopped uploading - hopefully it's a temporary break. I would love a video on your response time setup, explaining all the hardware and software so that we can do our own measurements at home.
I'm still here! I just don't have anything to review at the moment. But if you want to see more detail about how I'm testing, check out my 10% to 90% GtG video, where I cover response time testing. In that video, I go over exactly what I use to measure the screen. It's really just a micro-controller and a light-to-voltage probe, but the set up is a little bit complicated.
Would it be possible to revisit Apex Legends purely for Nvidia Reflex? They now added an in-game option to either disable reflex, enable reflex or enable reflex with boost. It would be very interesting to see if Reflex improves the game performance and response time or if you are better off leaving it off or on boost or normal reflex.
im looking for this exact answer.. i cant even tell the difference, im wondering if its worse to have enable + boost reflex while playing apex legends
Wow great work 👍
Awesome. Can you test Reflex on and off and ULLM on and off in cases where the system is not being GPU bottlenecked?
Reflex wins because it allow the game to run at higher FPS as long as the GPU utilisation < 100% whereas ULLM caps the FPS just below the monitor's refresh rate. Not to mention if both are enabled, Reflex overrides ULLM according to Nvidia
@@ADHD_Junkie No, I mean compared to both being off. I’ve seen some results before that show slightly less lag with Reflex off.
I wish we had a test with Valorant, I wonder what would have been the results
Great vid, new mouse? looks like it's changed
Cooler Master MM720. It weighs almost nothing... I love it.
Awesome job
After Battle(non)sense latency video's i have advised people to have 90%-95% GPU utilization for twitch shooters 🐱👤
Can you do a video on vulkan vs directX11 and opengl?
Can you answer me should I enable latency mode in cvar for destiny2 ? if yes then 1 or 2 should I also limit fps to 200 or 240 in 240 hz monitor ?
Thanks
I heard Quake Champions might hve some kind of anticheat that cause issues with external limiters (atleast if they have overlay). this was just after they released Reflex.
great vids! curious if you tried 144hz with elmb sync on v sync off with the asus vg27aq? it doesn't flicker at 144hz but does when set to 165hz... i have the asus vg27bq is a tn panel but faster response time 0.5ms... any thoughts if these monitors perform the same? thanks!
is it possible to add a GPU utilization % bar to Rivatuner or Msi Afterburner to set a limit for GPU usage?
I have an unrelated question and not sure if you know the answer: does UE4 have an internal engine limit of around 1000 fps? I tested it just with a test scene (self emissive cube, no light) in forward rendered mode and it seems to bounce off around 1000 fps or 1ms frame time. In Unity I can get around 2500fps with a similar test scene. Further in UE4 my GPU doesn't seem to even break a sweat whereas in unity at 2500 the thing spins up like a mofo. Could just be CPU bottlenecked.
I'm now on a mission to write custom openGL, vulkan, and DX 11 and 12 test scenes to see if I can find what can run the fastest. Using an RTX 3090 I managed to get for a wallet destroying sum of money lol.
I've been using the internal cap for 1000 FPS, but uncapped, it hovers around 1100-1200 FPS on my 7700k at 4.7 GHz. I don't think Epic was really thinking about optimizing for 1ms frame times, so there's a lot going on in the background, even on simple scenes.
If you're searching for something that runs at 5000+ FPS, check out github.com/NotThat/Benchmark-DirectX9/releases
Grab DX9.Black.White.exe. The guy who did these, NotThat, was doing something similar as well, writing DX11 and DX12 "games" to see which was the fastest.
@@ApertureGrille Thank you for the info and resources!
10:25
I might have missed it, but what is the reason that the same framerate for different GPU usages results in different latencies? It's a big deal!
It really good if you test nvcp limit and rtss limit in quake champ!
can you test the Pixio prime px248 type S please
If you're interested, could you investigate how in a game like csgo, blurbusters and some 240hz monitor owners recommend you turn on gysnc, and turn on v-sync in the nvidia control panel. Then turn off the v-sync ingame, and capping the fps 3-5fps lower than how much hertz the monitor has. This forces the game to always utilize g-sync, and won't have the stutter of the v-sync when the fps crosses the 240fps threshold, thus there is no additional input lag while using gsync.
I cover a bit of that in this video:
ruclips.net/video/8ZRuFaFZh5M/видео.html
But I'm not quite sure why people recommend setting V-Sync with the nvidia control panel rather than using the game to set V-Sync.
For instance, in UE4, using in-game v-sync and its internal fps cap works great. At 4:49 in this video, I show UE4 running that situation. V-Sync was set to ON for those measurements. The only reason the lag is slightly higher for those measurements than with v-sync off is that we aren't getting partial, torn frames.
Otherwise, the internal v-sync setting has no impact on lag.
I'll see if I can take a look at why people are recommending using the nvidia control panel v-sync option. I've heard that before as well, and I can't think of why it would make lag better.
@@ApertureGrille I think it was more on the lines of "the correct way to use g-sync". People still argue over what's best for a game like csgo, purist say that the best is still pure uncapped no syncs whatsoever to push as much frames for absolutely lowest input lag. But then, those that recommend using g-sync argue that input lag is either neglible or unaffected and the benefits or reduced motion blur and/or a tear free experience outweigh the cost of capping the fps. I've seen it done for other games like apex and overwatch. They say people could still argue the former at 140fps at 144hz, however it gets more interesting at 236fps at 240hz, I'm even curious to see how an overclocked panel like the asus can handle it at like 276fps at 280hz.but yeah, most of it are just claims of this and that, no one really has good numbers for comparison or the right implements to make a proper conclusion. But from what i can gather tho, at 360hz, i think there would be no argument for keeping g-sync on, and input lag would still be at comfortable competitive levels.
❤
If this was stated in the video, forgive me I didn't catch it-
is there a reason NOT to use reflex when available?
For the games that have it, no downsides!
How you get 220 fps in Apex with 1070ti. I have the same card and I struggle to have stable 120. When I'm fly in the ship over the map I get below 100 fps. And I'm already running at low and even went to 1080p on my 1440p g-sync monitor.... what do I do wrong?
I used to get more with a 1060; it's your CPU most likely
Hi a5hun, love your detailed analysis. Btw, have you ever noticed a horizontal line flicker for a millisecond (I'm talking so quick that you question whether you're imagining it or not) on your Legion y27q-20? I've had mine for a little over a month and at first I thought I was seeing things but lately I'm starting to think it's an actual issue with the display. I tried disabling gysync and lowering hz to 144.
Nothing like that. I've got it sitting on my desk right now, and I've been using it since Lenovo sent the review unit in December.
That's a hard thing to diagnose; there was a point last year where NVIDIA's drivers were causing flickering artifacts like that; I was seeing it happen on my VG27AQ, but they eventually fixed it. So it could be the driver or GPU memory or the DisplayPort cable.
@@ApertureGrille Realized I'm using an old driver. Will see if updating will fix it. I also have the vg27aq as a secondary monitor and it doesn't have this issue.
Nice video mate!
Just got a quick question, which 1440p 144/165hz monitor would you recommend currently ?
I haven't tested the LG 27GL850, but from other reviews, it should perform about the same as the Lenovo Y27q-20 and Dell S2721DGF. They all use the same nano-IPS panel, so you're mostly just paying for different industrial design and slight implementation differences. I think the Lenovo and the LG both have working sRGB emulation modes, which the Dell lacks, but you might just want to get whichever you can find at a decent price.
Can you test Overwatch 2 ? I wonder if i should let fps go to max with 100% utilization and enabled reflex boost or should i cap at 116 fps, gsync, low gpu load.
Can you do a test between using the Rivatuner FPS cap and in game FPS caps in terms of input lag?
Try this: ruclips.net/video/8ZRuFaFZh5M/видео.html
At around 7:41, I've got the full chart with NVIDIA's FPS limiter, LLM, ULLM, and RTSS. All the external limiters perform a bit worse than UE4's internal limiter.
So am running apex legends at 240 FPS, should i have reflex plus boost on or off? I have 3080Ti
ayo can you review a setup i wanna buy some day and can you spot out any flaws? i could reply with the setup or anything i should buy instead of one thing?
I wonder why you tried the Steam version of Apex. Doesn't the Origin version allow you to toggle NV Reflex on/off with a launch option or did I miss something?
Didn't know that, I was going off the NVIDIA Reflex page for Apex. Weird that it doesn't work on the Steam version.
@@ApertureGrille I could be wrong though, don't have a way to support my claims. I do know that I'll be downloading Apex on Steam, thanks for making this video!
please review the BenQ XL2546K
Hi.
What is the best setting please for warzone:
G sync: on
V sync: fast
Latency mode: ultra
Cap frame rate to: 350 fps
I have 3090 GPU, Intel I9 11900k CPU and 360hz asus monitor GSync
Please give me the best setting.
Thanks so much 🙏🏽
@a5hun do you have a discord account or server?
Nothing like that, but I do respond to comments. :)
should try GSYNC+VSYNC on with minus 3 fps cap lower than your monitors refresh rate..
V A L O R AN T. also nice mouse i cant believe no one uses it
Does this mean, that limiting FPS through in-game limiter is usually better than out of the game limiter?
Yes! At least if the game has implemented it properly. There's a chart here: ruclips.net/video/8ZRuFaFZh5M/видео.html
That shows a comparison of UE4's internal limiter, which is excellent, versus NVIDIA's control panel limits, ULLM, and RTSS's limiter. Internal does best, and the external caps are all roughly the same.
so it helps on apex?
I just found your channel after getting a 360hz monitor, do I need to worry about 100% gpu usage at these frame rates and is there an easy way to limit my gpu usage to say 90% as I am sure I am maxing mine out trying to get 360fps.
My monitor has Gsync
I think the Destiny 2 chart at 16:10 illustrates the high refresh rate behavior the best. At 220 FPS, running at 100% GPU use only increase the latency from 14.1 ms to 18.8. That's not so bad, and the difference, I think, would be even smaller at 360 FPS. Would anyone notice that 4.7 ms? Not sure.
With modern games, your framerate will naturally fluctuate based on the scene complexity, but I think it's a good idea anyway to set an internal framerate cap to something that'll keep you away from that 100% case most of the time. Even if you have to cap to 320 FPS, that's only a frametime difference of 0.347 ms compared to 360 FPS, whereas running at 100% use at 360 FPS might increase that by five times.
I know it's a bit irritating that you can't use all the GPU you've paid for without getting a lag penalty, but that's what we've got. Internal caps work best.
I'd also recommend using G-Sync. The lag penalty over V-Sync off is not that large, but you get a massive benefit with the reduction of screen tearing and judder. VRR is amazing, and I always recommend people use it... unless you're playing CS:GO at 700 FPS. :)
Brilliant thanks for the quick and clear reply.@@ApertureGrille
@@ApertureGrille I thought you might like to know my results (although I am about as novice as novice gets), with my gpu at about 60% usage and my pc running between 290 and 340 fps in siege with gsync on, I am able to use the nvidia reflex monitor built into my monitor and my pc+display latency results are normally between 10-12ms.
@a5hun quick question. Does this mean ullm does bad things when you run an old game at it's max capped framerate? I'm thinking about QuakeLive which doesn't use 100% of the GPU and CPU. For csgo I use ullm with my 6600k at 4.5ghz and 1080ti (1080p smaa 8x and aniso 16x shaders on low and textures on low) on some new maps I hit 90 fps and I really feel ullm helps reducing input lag.
ULLM only works on DX11 titles, so unfortunately it won't do anything for OpenGL/Vulkan/DX12 titles, and that includes Quake Live.
I haven't actually tested Quake Live for lag, but QL is a weird one because it's not as adaptable as other games; it tops out at 250 FPS, and the framecap only caps to integer divisors of 1000 (so 1000 / 8 = 125). For QL, I'd just turn V-Sync off and run at 250 FPS. You'll get tearing each frame, but there's not much that can be done.
How are you getting such low framerates in CSGO? Your PC is pretty great. I don't think ULLM will hurt in CPU bound scenarios, maybe 1 ms or so, but it's only really useful if you're 100% GPU bound.
@@ApertureGrille yeah I have no idea how I am getting so bad framerate in CSGO. Is there a benchmark that I can use to test within this game? As for QL, I use a reshade motion blur shader that completely remove screen tearing at the cost of 0.4ms GPU time. It feels great.
I just had confirmation from khronos group (devs of vulkan api) that it does affect dx9 10 11 & opengl.
@@jeanmoreau4868 I'm glad you asked. I just checked the NVIDIA site and they only say it does not work in DX12 and Vulkan.
bind "F1" "toggle fps_max 60 120 144 180 240" in autoexec next time you wanna test fps limits in apex
But how much of that does actually matter, when human reaction and other factors like internet latency and the consistency thereof is probably more wonky than 3ms difference (3ms taken from the QC graph as difference between 420fps max and 240fps cpu bound). AFAIK human reaction, depending on what happens i.e. expected/prepared events vs. unexpected/unprepared events and mental stack, the variance is probably WAY bigger than that.
Second thing is how certain interactions are animated i.e. how many frames and which action are possible (I'm coming from a fighting game, if that isn't already clear ;-) )
Like the time a shotgun or a rifle is reloaded is going to be the same in seconds, but no in frames, obviously.
While I certainly agree to never play GPU bound at 60fps and rather 120 or even 240 if possible. Further bigger improvements, especially in the beginning, are easier done in the game i.e. getting better, decision making, etc. than adjusting anything else outside the matches boundaries.
Otherwise there really isn't anything to take seriously, if the gap is going to be that big with highly different machines.
BUT, that wasn't the point here. I know.
Good points. Chasing down a 3 or 4 millisecond reduction in latency isn't particularly worth it if it costs $500 or you have to reduce the rendering resolution so low that you lose visibility.
Check out this program flood posted on the Blur Busters forums that lets you A/B test for lag:
forums.blurbusters.com/viewtopic.php?f=10&t=1134
I think most people would have trouble correctly determining any latency below about 20 ms.
Would capping your GPU power limit to 95-97% in MSI afterburner fix the 100% GPU utilization input lag?
No, they are completely separate things. You need to cap FPS in each game individually.
a quick question: why you didn't use frame rate limiter slider in warzone to prevent 99% gpu scenario?
Because I was testing with a QHD monitor, even with the resolution slider at 50%, my 1070 Ti isn't fast enough to prevent that 99% use when measuring at 220 FPS, so I dropped the CPU-limited (or capped) results.
It's not ideal that I'm testing with a 1070 Ti, and I was hoping to get something more powerful, but video card prices are out of control right now.
A few people mentioned that you can adjust that rendering resolution lower by editing the config files, and I may try that for the next video. And Apex Legends recently got an update with a working Reflex ON/OFF toggle, so I can test that as well.
@@ApertureGrille No my point was not resolution slider. You should use framerate limeter inside warzone. For example if your gpu can render 220 fps with 99 % load just set fps inside warzone to 180 fps and load on your gpu decreases. Look at the video the fps cap is unlimeted.
@@alikalhor77 Oh, I understand! For the testing, I was trying to hit three consistent framerate targets amongst all the games: 60 FPS, 120 FPS, and ~220 FPS. That way I could make a fair comparison at the end.
For all the other games I tested, I was able to reach ~220 FPS without running into a GPU bottleneck, but with Warzone, that same 220 FPS was hitting the GPU limit.
So instead of capping at something lower like 180 FPS, I decided not to include the CPU limited tests; that would have had me comparing 60-120-220 for every other game with 60-120-180 for Warzone.
180 FPS vs 220 FPS is only around a 1 ms frame time difference, so it wouldn't have been that unfair, but I think people might have gotten mad at me if I compared the two when they're 40 FPS different :)
@@ApertureGrille very good content all around. the video from battlenonsense suggested that reflex works exactly the same as not gpu bound scenario in terms of latency but your findings shows difference between the two . I think it's about implementing and and result will be game dependent. He tested overwatch and fortnite I think 🤔
@@alikalhor77 I think that difference may be down to how absurdly low I dropped the GPU load. For instance, I don't think anybody is going to be playing Destiny 2 at 25% resolution scaling. At that setting, my GPU was only around 38% load.
So while that's not a realistic use case, I think it does help illustrate how quickly the game engine can react if it's not being limited (or barely limited) by the GPU. It's kind of like doing CPU tests at 720p.
Apex Legends get horrible frametime issues over 190 FPS, not sure if that matters. I wish I knew that before buying a 240 Hz monitor.
Awesome video though I'm still questioning my settings. I have a gtx1650 4gb on ryzen 7 2700x .still don't know if I should have reflex off or on or on+boost on warzone
For games that have reflex, there's no downside to turning it on. Reflex reduces lag when you're GPU bound, but doesn't hurt when not. But even still, you should try not to hit 100% GPU use.
With your 1650, that might be hard because Warzone is pretty heavy on GPU, but try to turn down settings and cap the framerate to something you can reasonably maintain.
I have always wondered if relex increases the amount of time it takes to see something on the screen for you to respond to. It seems really hard to test that though. Like for example with reflex off would you see that guy coming around the corner sooner than you would with it on? Maybe not, but I don't know enough to say really.
Make videos with v sync on?
GTA online with V sync please.
Actually the golden setting is G-Sync + Vsync.
Go watch that Battle(non)sense video that has his test results and view on it.
So is ULLM only useful at low frame rates then?
Low-ish framerates and high GPU use! For DX11 games, it does "OK" at 120 FPS. In Destiny 2, ULLM cut the lag from 41.2 to 35.9 ms in a GPU limited scenario; that's not a ton, but it certainly helps. But you're right, it does best at ~60 FPS.
Generally, ULLM is a good idea if the game you're playing doesn't have a good internal cap, so you keep bumping into that 100% GPU wall.
@@ApertureGrille Ah okay that makes sense.
Are there any cons to using it?
I'm a QuakeC main and get a pretty consistent 200+ fps (using Rivatuner to cap at 280fps)
thanks!
@@FabFeline I should have tested QC, but for Destiny 2 at 200+ FPS, it looks like it may hurt slightly... maybe less than 1 ms. Not something I'd worry about.
@@ApertureGrille Okay good to know, thank you! great vid
Can you use ULLM with reflex or does one overwrite the other?
Reflex overrides ULLM according to Nvidia
If Reflex is enabled, it'll over-ride ULLM.
CRT monitors: "look what they have to do to mimic a fraction of our power!"
he kinda sounds like Patrick Bateman
Btw you can disable reflex in apex via config file
When the video went up, this NVIDIA page: www.nvidia.com/en-us/geforce/news/nvidia-reflex-apex-legends-out-now/
said that the +gfx_nvnUseLowLatency cvar didn't work due to a bug.
The season 8 patch came out only a few days after I posted the video, and they finally added Reflex in the menus! Haha! Just in time, Respawn! :)
I need to hop back in and re-test it now that it can be turned off. And people were also saying that Apex behaves badly above 190 FPS, so I want to check that out as well.
@@ApertureGrille People may be right, I can play with 160-190 fps, but there is some stuttering on my system in 1440p (10700k, rtx 2070, 3800mhz ram, nvme ssd). And sometimes I get huge fps drops to 40-50 fps in some scenes. If fps is stable 140 the input lag feels inconsistent.
@@ApertureGrille Please, if you will do comparison in R6 Siege, make comparison of versions DX11 with fps limiter and ULL vs Vulkan + Reflex and limiter. For my feelings DX11 version seems to be more responsible than Vulkan with reflex, which was surprise for me.
the advice here is weird cause...if you disable vsync in most games, your GPU will run at 100%.....but at the same time you're saying NOT to run your GPU at 100%........so I guess set a frame cap that's over 200fps but not so high that you'll max out your GPU usage...?
That would work, but there are other ways as well. For something like Quake Champions, if you drop the render resolution a bit (not to the crazy 25% I was using), in real gameplay like a multiplayer match, you'll hit CPU limits instead of GPU limits.
If I drop my rendering resolution down to 80%, in bot matches, I'm usually hovering around 150 FPS because of CPU limits, and the GPU is sitting at 85 to 90% use.
Conclusion is: Give me all your FPS Nvidia & AMD. Thousands of them/second while you're at it.
what is point in measuring new frame which is only at bottom, playing game it is useless information, what is middle is what you aim, so only when this part of screen is refresh it gives you useful information.
Obviously very late but HOW in the fuckin world is your gpu running at 35 C at 70-80 usage when running that mess of a game named "Quake Champions"??
You got your pc outside in Alaska or what homie?
Pretty hefty undervolt on the GPU (0.881V), but I also have the Strix 1070 Ti running with liquid metal. The liquid metal makes a huge difference, so much so that any GPUs I have that have a nickel-plated heat spreader are getting the LM treatment. The 7700k also has LM under the heat spreader.
you havnt even scratched the surface . QC has very strange input lag that effect everything mouse. movement. desync of opponent model. the worst shooter from technical side i have ever seen