Thank you for shining light on this issue! Something else I've noticed is that on older machines you can greatly improve smoothness and frame times with a frame rate limiter. Works especially well with a variable refresh monitor...
PhilsComputerLab. Hey, howzit going Phil. Nice to see you here. You are absolutely right. I always limit my frames, even on my new rig. I don't want my CPU going anywhere near 100% CPU usage (I don't even want it near 90% to be honest) so the less the better. You are letting your CPU breath as it does not have to feed the GPU as fast as possible. A variable refresh monitor is a fantastic companion when doing this (and in general). On older builds its a godsend
@@philscomputerlab Exactly. It really does lower stress on all components and leads to happier system and nicer all round experience. It is something that is not mentioned by enough people.
@Gotta Go Fast You're right bud. Adoredtv has a great video covering Radeon Chill. Nvidia has just added a feature in its control panel to manage your frame rate too now called "Max Frames" (although it has none of Radeon Chills other features). But MSI Afterbuner is excellent too.
Even If his math was right which is technically not his argument is a non sequitur because if part of something is small compared to the whole that doesn't mean it will not have an effect. If it did 400 parts per millions of carbon dioxide would not effect global temperatures.
If you wanted the short answer I don’t think RUclips is the right platform for you to get your info! But I guess blindly accepting “yes” and “no” answers works for some people
Honestly you need to be a teacher. The way you explain things makes so much sense for anyone to understand. And your detail is worth noticing. Keep up the great work.
I've settled on 1% low being important, 0.1% much less important. Hardware unboxed uses average and 1%, Gamers nexus uses all 3, and includes error bars which shows the variability. I'd say 1% or 5% lows might be useful to show the typical slowest frame in a given second. After that, some metric that can capture severe stutters or hangs would be nice. I remember seeing when RDR2 came out it would have some multi- second hangs. It that was one frame I'm not even sure it would show up as 0.1% low stats. Maybe something that looks at frame time and will highlight frames that take 2× or 3× longer than the average to see if they are an issue at times other than loading screens.
this is legit the 3rd video that i watched to learn what 1 and 0,1% low were the other 2 were too complicated explaining frame times and stuff thx for explaining it simple
1% lows are the main thing I always look at in benchmarks. I never understood why the .1% lows were there, because it's literally a pointless factor. I would rather have a CPU that averages 100fps with 80fps 1%, than a 110fps CPU that gets 70fps 1%. You can easily pick up a CPU nowadays that run games flawlessly such as the 2000/3000 Ryzen and 9700/9900k.
But would you rather have averages 100 fps with 80fps 1% and 30 Fps 0.1% than 110fps 70fps 1% and 60 Fps 0.1%. The first example will result in 33 ms frametime every 10th second which is a stutter. A lack of memory can do that. I am not saying there is no case to be made for abolishing 0,1 (0,1 is noisy hard to replicate) but 0.1 is not a completely useless metric.
This was incredibly useful. I had a 4790k and I was looking at going to a 3700x or a 3900x and I was disappointed that the average frame rates were so similar but as you said its that 1% low where you feel the difference. games are much smoother; start up is much faster etc. Just mirroring what others have said you nailed it on the head.
I went from an i7-3770k to a Ryzen 7 3700x and the difference is massive, all the little stutters and slowdowns in games that I never managed to get rid of even by lowering the settings in the options menus are now gone. And I agree 1% low is just as important as the average framerate, but 0.1%low measures tend to be way too all over the place to be truly useful.
Yeah. I have a build I will be working on later this year and the most important thing to me when it comes to CPU is how well it will handle frame pacing now and in the future. You are one of the few channels that I've seen that actually focuses on this aspect and it's greatly appreciated!
Tech Deals, you are absolutely right. My frame delivery consistency improved considerably going from a i7 2600 to a r7 3700x (as i expected it to). And yes, the upgrade was to improve the consistency between the lows and average frame rates taken as a whole and not my max frame rates. PS: I think you are on the right track in experimenting with how you benchmark and the way those results are presented to us. If anyone can figure out the best and most reliable way to achieve this, its you.
Same here going from 2500K (4.7Ghz) to 3800X (4.5Ghz). The frametimes are much more consistent so even 60fps feels nicer. What GPU are you using and did ytou change you GPU too. I had and still use GTX 1080. Honestly the platform upgrade pushed back my GPU upgrade considerably. No point in pairing 2500K with anything stronger if the frametimes are a mess.
@@Raivo_K Absolutely. That was a massive upgrade for you too. Great pairing with that 1080. Running a 1070ti at the moment and its like I got a whole new GPU with the new platform. But here is an interesting tidbit. I paired the i7 2600 with a 1050ti for a secondary rig and in newer more demanding games it bottlenecks the 1050ti (Detroit: Become Human) being the latest. Even in older games like Shadow of War. When I swapped the 1050ti into my R3700x rig the mins went up and turned it into a GPU bottleneck. I believe its the per core performance and the DDR4 3200mhz (tweaked timings) that help. The 1050ti is really low on bandwidth. Anyway, what games are you playing and are showing the biggest improvements?
@@joeyvdm1 I don't play that many games at the moment. I was playing Mass Effect Andromeda multiplayer with friends. There i noticed the biggest improvement. Frostbyte engine loves 8 cores. Even when booting to single player my CPU was at 100% constantly for minutes before it stabilized and fps went to mid 50's. Now i get faster loading, no waiting and fps is 90+ straight away. I believe all Frostbyte based games should see a nice per uplift. Also Quake Champions become much more fluid.
@@Raivo_K Absolutely agree. Multiplayer gaming and Frostbyte engines make great use of 8 cores (and i suspect more, to an extent, if you've got them). In single player gaming (Shadow of the tomb raider, Jedi fallen order, Red Dead 2, Detroit Become Human, etc) I have gotten major upgrade including load times (as you mentioned). EDIT: All round system performance and snappiness is fantastic too. I do lots of video converting and I can encode with a gazillion tabs open while watching youtube and working in photoshop and it is still as responsive as if I was doing nothing.
Frame time variances as well as fps variances are the most important parameters. They are achieved by calculating the derivative of those to parameters. That's what human vision is more sensitive about. I much prefer a constant 60fps experience then a 60 to 120fps, all over the place, one. If the fps variance is slow and gradual the brain adapts to it easily but if it's changing by the second it is perceived as stuttering. That's very distracting and completely destroys immersion for me. That's the reason why I block my maximum FPS, with rivatuner, to a number that alows me to have the best and smoothest experience possible even if that number is well below the average fps achieved by my system. That's a number between average and 1%fps and how close to 1% really depends on the game.
I think 0.1% low has its uses but the limitations are significant. It's best on built-in benchmarks and will often reveal things like microstutter. Consistently low 0.1% numbers are usually a legitimate issue and not just some random windows did something in the background. At the other end of the spectrum, an extended playthrough with respawns, saves, loads, etc will render it worthless. In the middle, on a stretch of playthrough that doesn't have those kinds of interruptions, bad 0.1% numbers are often indicative of some kind of real performance issue that merits further investigation.
I don't know what to think. My first reaction is to ask, would using standard deviations from the mean be better? Instead of using the mean, would using the medium and standard deviations from the medium be better? Your comment, it's better to have a consistent frame rate of 30 fps than have half your frames be 30 fps and have your frames be 90 fps, is very interesting.
Thx for this really interesting video! I totally agree with you. I also benchmark with Afterburner on my own machine. I just let it run in the background, while I'm playing. Sometimes the 0.1% is 0 or 1, although the game ran really smoothly for an hour. Maybe I just died several times or reloaded the save game. Steve form Hardware Unboxed never uses 0.1% lows. Maybe you know his channel. In my opinion 0.1% lows only make sense if the benchmark doesn't take any longer than 30s and is repeated several times to eliminate a potential outlier.
Thank you for being the only person I can find online who shows just how extreme 0.1% lows truly are. I've now changed my benchmarking analysis to only include AVG and 1% lows in future. Why do you think 3 and 5% lows might be better?
how do i keep the 1% fps stable? i noticed that when i play at 60 fps, 1% many times drops even to 30..making a fps drop..how do i keep it stable? or as high as possible?
Thanks TechDeals for tackling this issue. Average FPS is just that -- an average. And trying to sift frametime anomalies out of a subset of an average is froaght w/issues. Back in the day Scott Wasson at TechReport was championing frametime analysis, then PCPerspective experimented with some different graphing techniques, and GamersNexus continues with an occasional frametime plot. Sadly the more advanced analysis never became the norm.
I get on my rtx 2070 super and ryzen 5 3600 and 16gb ram 0.1% lows on paladins to 25 fps from 400 and i feel the stutter all the time how can i fix this temps are at max 60C
I'm trying to understand frametime stuff better, so someone please correct me if I'm wrong. I think I disagree with the video on the subject of 0.1% framerate being useless. One of the points that the video makes is that 0.1% low is a bad measurement because it's such a small amount of frames it doesn't really tell you much about the whole experience. His example was that 0.1% of 60fps gameplay over 10 minutes is only 36 frames which is about a half second, which makes it sound like somewhere in the course of 10 minutes of play the game will freeze for half a second. That's not a big deal. But I understand it differently. If you're playing at 144fps, then the 0.1% low in 10 minutes of gameplay is about 86 frames. Rather than experience a half second stutter once in 10 minutes, that's 86 separate times that you're going to experience that 0.1% low. If the 0.1% low is bad enough, that's 86 times that you go to make a headshot and the game stutters and you miss the shot because of it. I might be completely misunderstanding this and if so then please let me know.
How much does capping the FPS/enabling Vsync affect frame consistency and stutters? All of your CPU tests are pushing the absolute maximum FPS and pinning the CPU at 100% (which naturally leaves few resources for system tasks), but many people play with Vsync or a cap and never get close to maxing out their CPU. It would be interesting if you could test frame time consistency between locked and unlocked frame rates to see if leaving some headroom could keep older CPUs going just a bit longer.
what would be a good 1% low statistic? my 6700xt always shows a numeric value of 2, I thought it was supposed to mean that sometimes (if my average fps is 120) then every once in a while my fps should go down all the way to 2? that shows extremely bad to me but I always have a smooth gaming experience. Then again I don't know much, any help please? is that a good statistic or not?
Gamers Nexus touched on this superbly in their 9600K review. The main thing I look for is frametimes, and if I see a spike on the frametime chart (which only occasionally show up in Afterburner's framerate graph), then my flags start going up.
Finally!! A video that explains 0.1% and 1% lows in a way I can wrap my head around. The key wording he used was "1% low means 99% of the time your FPS will be above this number". No other video (gamer nexus, RandomGaminginHD, bitwit etc....) about 1% and 0.1% lows took the time to elaborate on this one specific but important explanation. Great job!
Ive just upgraded from.an i5 4690 to ryzen 5 2600x keeping the same gtx 1060 and the experience is soo much smoother, yes im not really getting any more fps but it feels much nicer
GREAT Video!!! Between what I have learned from you and Digital Foundry I've been heavily focused on the 1% lows and .1% lows over Average FPS and learning to tune my graphics settings and cap frame rate to get a buttery smooth game play. I think a lot of us gamers (which I'm very much guilty of) are/were too focused on Average Fps numbers. This is why I upgraded from the 9700K to the 9900K because while the 9700K has the potential to OC to 5.2-5.3ghz the FPS increase does exceed what you would often get from a 9900K but what you don't get is the Hyper Threading to smooth out those 1% Lows for buttery smooth game play. For the record if I didn't own a 2080 Ti or plan to upgrade to a top tier card in the near future AMD is the answer because the cpu performance is a great fit and won't bottleneck the gpu but if you do own a 2080 Ti and play at 1440p you really should be pairing that gpu with an Intel chip to get the most of that gpu. Otherwise why buy the card? You would only gimp the performance to almost a near 2080 Super experience :/ I digress - JayzTwoCents ; in regards to the .1% lows I think you are probably right, its the 1% lows that seem to carry the most weight. Awesome channel you have hear man. I love the fact that you don't just review hardware and give the boring benchmark graph charts but you do lots of testing and demo what you are testing. This is how we learn, so good job man keep it up.
The only way 1% lows matter is if you calculate them via a circular buffer that constantly updates, so low numbers dont saturate the entire benchmark. You then calculate the amount of time certain 1% lows occurred during a set time period and possibly develop an average aswell. So that would imply a 1% low frame rate thats, on average, either closer to the highest FR or not.
so is not possible for the Ghost Recon example to have 10 fps in the 0.1%? that would be a couple of ugly skipped frames or even periodic stutters. Some skips and stutters while very annoying wouldn't appear in the 1%, unless they are a lot of them, but 1 ugly skip every minute would be awful. Just 4 runs show some variance but I don't think is too bad, just mean more runs are needed, for it to be useful, that "evidence", tells me not that 0.1% is useless but that 4 runs is too few. That data doesn't show randomness as you suggest, far from that. Of course it may be a better metric to shows frametime consistency, in that case great, but I don't think there is any evidence that 0.1% is useless.
Great video explaining the uselessness of the 0,1% low. May I suggest videos on combinations of CPU and Video cards that would pair together to give the "best" experience at 1080 and 1440 and 4K gaming. Or have you done this already?
very important question,,,,, if your 1% low is lower than your monitor hz is that bad then? or is it not noticable at only 1% of the time out of 99% being higher than the monitor hz
Just to clarify, you can get better fps, or better timing, or faster frames by upgrading. I went from avg 50fps on an RX 590 with an FX-8320 on high settings in most triple A titles. I bought a 3800x and kept the same gpu and I now get 60-70+ fps avg with no lag or issues. So you can get better game performance by just updating the CPU. I'm not sure if you said what you said slightly off or if I misheard you. My apologies if I misheard you, but I hope this info helps. Also, it's not alot of performance gain, but it can be just enough to push you over on fps. The misses rocks an FX-8350 and I know there's gpu performance just sitting on the table, so I'm gonna try to get her on Ryzen for that same reason. Edit: Rewatched video. I heard you wrong. And you covered all I said above. Thanks for the video!
One possible way to test is to cap the max fps at say 60 fps. Then the average fps becomes meaningful as extremely high fps wont be able to cancel very low fps and hence will accentuate performance dips. The closer they can get to the cap (60 fps), the better the card is.
Tech, sorry to comment before watching the video, but can you please rename the video with the word "very" in front of "useful" just for the sake of importance. Be assured that I'll watch the video when I'll finish work, keep up the good stuff for us, you're awesome!
Thanks! Very interesting! You mentioned 3% low etc. Wonder what info afterburner dumps from the benchmark and if you can calculate it with Excel using the perceintile function?
I love your work and how you truly explain everything. Before I ever make a purchase I always go over your videos on it. And I have been meaning to tell you, love the new intro. Keep up the good work.
I'm glad you are looking at the commonly used benchmarking system today and identifying it's flaws. I agree to omit .1% since it's not significant to the experienced gameplay. I will suggest a benchmarking system which makes a graph of fps versus percent to show how 'often' you can expect a certain fps. A narrow bell-shape will be better, a flat dome-shape will be worse. With this, one can see how much the fps varies, how often the fps are under accepted level, and get a better view of frame-pacing.
Hi, I need some advice regarding of gpu, I was planing to get new gpu, and at first I wanted to buy rtx 2070 super, but after recent events, I started to consider buying rtx 2060 super, the price for 2060s is ~480$ and 2070s ~630$, so here my question is it worth it to pay extra 150$ for 2070s, or 2060s will be enough for 1080p maybe some 1440p for the next few(5) years ?.
it's like riding as a passenger in a moving car situation 1 passenger = a driver drives at 120mph and sudden brake to 40mph in a second and goes up again to 120mph next second situation 2 passenger = a driver drives at 60mph and sudden brakes to 40mph in a second and then goes up to 60mph again next second i would choose "situation 2" passenger experience rather than "situation 1"
I agree with you, Tech. When looking for at-a-glance performance indicators for a given set of hardware, I always assign 50% of the score to the average frame rate and 50% to the 1% Low. Both of those numbers are very useful.
Oh and adding to my other comment: The apparent difference in accuracy between 0.1% and 1% is merely a function of how long your benchmark is. It just so happens that you (and most people) tend to run benchmarks for long enough that 1% of the sample is still meaningful but short enough that 0.1% picks up "noise". If you ran the same benchmarks for 10 times as long your 0.1% results would become as meaningful as your 1% results are now. In turn, if you ran shorter benchmarks you might have to give up the 1% category. Assuming that the "noise" really comes from outside the games it should be constant, at least on the same PC with the same programs running, so we could calculate a precision cutoff depending on the length of each benchmark. So for longer benchmarks we could go all the way to 0.5% for instance, and for something like the Far Cry 5 benchmark we might stick with 3% or whatever.
Should I cap my frame rate to the 1% lows? or around the average framerate for the smoothest gameplay? Specifically Esports games. For example on Apex I average around 80fps and have 1% lows of about 50fps.
Also appreciate this was a more bite-sized video! I love your videos but prefer 10 min to 30 min videos as like to watch youtube inbetween doing other things!
I mostly agree with your statements. 1% low is definitely enough when 0.1% is above a number where you notice lags and jumps. When the 0.1% value is 30 or even 50, then you'll never ever notice the moment when it happened. But let's say you measured a value of 10, I bet, you could tell me the moment when it was measured given the fact, that some things were moving in the game.
After finally starting to do all my benchmarks to include 0.1%, I immediately I realized how stupid it was to rely on this specific metric as a critical performance indicator. As a person benchmarking a system, it truly means nothing to another user who will have a different program load on their CPU. The ONLY WAY to selectively benchmark the minuscule window of “good” 0.1% for the sake of being able to show impressive numbers is to play in such a way that does not at all represent a typical usage scenario. I’m with you on this one. Great breakdown.
I totally agree with you a static frame rate or a frame rate with low variance is better than when it jumps all over the place. Especially with synthetic benchmarks, sometimes the frame rate may ideal around 25 fps than jump to 75 or more for a few seconds just to return to 25 fps. I wonder how many people realize that is just how the system behaves depending on how demanding the game.
Input lag will be better. Knew I wasn't crazy. My i5 6500 with a gtx 1070 completely ruins my gaming experience. Input lag is mad idc what fps i get. Is all of this linked to cpu usage in any way?
This was a very informative crack at a complex problem. Average fps rates are easy to understand but the truth of the gaming experience is more nuanced. Unfortunately, people have increasingly short attention spans and less and less time to consider complexities. Sound bites and summaries only, please! Sadly, this is the equivalent of information fast food.
.1% lows basically tell us if we can expect to a game to have severe hangs and snags. Its significance doesn't come from what it tells us about a framerate but rather what it tells us about a game's stability. It's certainly less useful than the other statistics considering that improving it is often much less straightforward than the others if it can improved at all, but it does serve a purpose.
How hard is it to get the standard deviation? Give me the average to show general performance and then something like the breadth of the 99th percentile. For example, one rig might produce 60 fps with 99% of the frames within +/- 15 fps while another gave 60 fps with 99% within +/- 20 fps. The gap between 1% lows and highs would indicate the smoothness, lower being better.
sometimes is useful talk about the gaming experience and not only numbers. I have a 9600k and his fps stats are high but it feels a little trash regarding the gaming experience. Waiting a motherboard for a 3700x and check if the gaming experience will be better. thanks for that video
@ZhekaTrololo It depends bud. If you are happy with the performance (and by your statement it sounds like it) then wait until it isn't good enough for you anymore. If you are asking the question because the thought hadn't crossed your mind before (because it is all good for you right now) then no. Wait until you aren't satisfied anymore. PS: I wasn't satisfied anymore with my performance in newer games, so I upgraded.
@@joeyvdm1 Yep. I had 2500K and while in older single player games it was totally fine it struggled massively in newer AAA and especially Multiplayer heavy games. So it also depends a lot on what you're playing and what GPU you are using. Im using a GTX 1080 so i always knew my 2500K was holding it back somewhat. Even at 1440p high settings but upgrading to Ryzen 7 3800X last year gained me about 25% more performance despite still using the same card. Especially multiplayer games went from stuttering or occasional hitches to buttery smooth. So with the old CPU i might have had 60fps but the frametimes were all over the place. 2600 has HT but i feel like even that is starting to show it's age. Newer CPU's don't add just cores. The IPC and thus single core performance also goes up nicely.
@@ZhekaTrololo No one knows your situation better than you bud. If you believe you have a better chance of selling your current rig to help finance costs for an upgrade, that is an incentive to do so. Okay cool. Let me see if I can help you. What is your current rig (GPU,RAM) and what would you like to upgrade to or have you no clue yet? Are you looking for suggestions?
I had a i5 9600k 5 ghz, the Witcher 3, which are not the newest game by far, sometimes made some strange stutters, just for a second, I hated when it happened and It was very annoying. And I started to monitoring my CPU usage, GPU usage, FPS, temps etc. and that moment, when those stutters happened the 6 core CPU did run up to all core 100%, but most of the case the CPU worked on a high precentage. I have Steam, Epic Games app, Origin app etc. running in the background all the time, also CAM software for my Kraken x62 AiO, sometimes I forgot close the browser. So I did some research on Reddit and they said to me go with a higher core count CPU, because that could be the only problem, I have an RTX 2080 + 32GB Corsair Vengance 3ghz CL14 memory, so I sold my Intel mobo and CPU and bought a Ryzen 9 3900x, becaues I considered to start making youtube videos again. Anyway all of these hiccups are gone, I have a much smoother experience, even though the avarage FPS dropped a 5-8 fps, but who cares, the 1% lows are elevated from 10 fps to 80 fps, that's something and I don't have to bother anymore the background apps, they can run without problems!
I'd love to see a new type of graph for frametimes. For game benchmarks where you can get a count of frames and a log of frame times, quantize them into millisecond "buckets" -- e.g. take number of frames that took over 500ms to render, then 250ms, 125ms, 66ms, 33, 16, 8, and under 4ms. Plot frame total on the X axis, and plot frametime bucket on the Y axis with slow frames at the bottom, fast frames at the top. Connect the plot points to create a curve. I'd call this a "FrameTime Quantization" or FTQ for short. "Good" results would have a sharp peak as high and to the right of the chart as possible. "Bad" results would have multiple peaks. The tallest peak would equate to a "median FPS" - e.g. a sharp peak at 16ms would be a "solid 60 FPS".
What I would like to see is the end of "fps". I believe the discourse should always be about frame times instead of frame "rates". Why? Because 100 fps can feel great (Battlefield 2) and they can feel awful (CSGO) so we learn nothing from that number. We need metrics that get to the heart of what we want to measure and that's how responsive games feel. To me the "1%" and ".1%" terminology obfuscates even further what really matters than the standard term "framerate" already did. When it comes to responsiveness we care about the attributes of EACH INDIVIDUAL FRAME, such as: Latency - how old is it, how long does it stay on screen? Synchronization with the monitor - is each frame complete or are we getting partial, torn frames? As far as I'm concerned ONE bad frame in 10 minutes ruins everything. If a game runs at 1000 fps for 10 minutes, except once a single frame stays on my screen for 50 ms the game has a performance problem. 1000 fps and bad performance at the same time! So why talk about fps? What matters is the responsiveness I can RELY on. If you go climbing with a rope that holds your weight 99.9% of the way up a cliff the final rating for the rope is 0/10, useless. You fall off the cliff and die. The 1% lows in frame rate or peaks in frame time get closer to the heart of the matter but really even they are too generous, that's where the desire for 0.1% numbers comes from. If the game fails to respond reliably at any one point then the game is overall unresponsive, no matter how many frames it pumps out before or after. True, reliable responsiveness is what we need to measure, and we need to figure out how to dismiss "noise" caused by Windows quirks or driver bugs or whatever, without losing sight of the ultimate goal. The ultimate goal is not and never has been maximizing the number of frames. When you look at how terribly inconsistent modern games are I also suspect that game developers have lost sight of the goal and are optimizing in ways that don't actually lead to responsiveness.
I always ignore .1% lows and disregard any benchmark channel who use that just to trash games they don't like, even if the average and 1% are fine and that .1 may just be a momentary stutter due to whatever reason.
well 0.1% might just be a buzzword created by the tech industry journalist to entice the viewing public when all it really matter is the credible reporter saying "this game has a superb frame pacing and frame delivery on xyz hardware meaning youll have a great and very smooth gaming experience." in my experience Ubisoft is the culprit in this hulabaloo when they released watchdogs 1 requiring more than 2 threads/cores to run smoothly as having less means 0.1% or even 1% low of 0fps meaning the game pauses from time to time to catchup and it totally breaks the immersion experience. sadly this is now rampant on most games. i also remember back then in gaming we never cared for slowdowns framedips etc, we just look at it like it is part of the maximum experience the game devs are trying to offer, ridge racer 1 on ps1 was awesome even if its just 320x240/30fps and dips occasionally while the arcade is rock solid 640x480/60fps with way better textures. we have been spoiled! imho we need a new display monitor technology that combines the best features of flat panels and crt's, as most of pc gaming buyers decisions revolved around that pesky hard to drive flat panel monitor (yes monitor dictates what gpu youll buy then the rest follows).
Thank you for shining light on this issue! Something else I've noticed is that on older machines you can greatly improve smoothness and frame times with a frame rate limiter. Works especially well with a variable refresh monitor...
PhilsComputerLab. Hey, howzit going Phil. Nice to see you here. You are absolutely right. I always limit my frames, even on my new rig. I don't want my CPU going anywhere near 100% CPU usage (I don't even want it near 90% to be honest) so the less the better. You are letting your CPU breath as it does not have to feed the GPU as fast as possible. A variable refresh monitor is a fantastic companion when doing this (and in general). On older builds its a godsend
@@joeyvdm1 And it reduces heat, noise and power...
@@philscomputerlab Exactly. It really does lower stress on all components and leads to happier system and nicer all round experience. It is something that is not mentioned by enough people.
@Gotta Go Fast You're right bud. Adoredtv has a great video covering Radeon Chill. Nvidia has just added a feature in its control panel to manage your frame rate too now called "Max Frames" (although it has none of Radeon Chills other features). But MSI Afterbuner is excellent too.
frame rate limiter always works better than v sync for me.
I'm in favor of dropping .1%, keeping 1%, and adding whatever measurement you feel best tracks percieved smoothness.
This makes more sense, I can’t believe No other RUclipsr ever addresses this. I’m really glad I found your channel sir
The most underestimated tech youtuber
o 3o with the brightest smile
I upgraded from i5-4440 to r7 3700x, man it feels day and night
3:21 Tech Deals giving example backs up by math
Me : New knowledge learn
Even If his math was right which is technically not his argument is a non sequitur because if part of something is small compared to the whole that doesn't mean it will not have an effect. If it did 400 parts per millions of carbon dioxide would not effect global temperatures.
You are completely right. Smooth gameplay is the most important thing. Perfect frame times especially.
Short answer,
1% yes
0,1% no
If you wanted the short answer I don’t think RUclips is the right platform for you to get your info! But I guess blindly accepting “yes” and “no” answers works for some people
So Gamers Nexus is wrong basically. Ok.
@@hhectorlector
1% low yes,
.1% low no
0.1% is useful, you’d notice this as stutter. If it not measurable by reviews, at least have a note saying stutters was seen.
No, because a % is based on an end total. Not a constant variable.
Honestly you need to be a teacher. The way you explain things makes so much sense for anyone to understand. And your detail is worth noticing. Keep up the great work.
I've settled on 1% low being important, 0.1% much less important. Hardware unboxed uses average and 1%, Gamers nexus uses all 3, and includes error bars which shows the variability. I'd say 1% or 5% lows might be useful to show the typical slowest frame in a given second. After that, some metric that can capture severe stutters or hangs would be nice. I remember seeing when RDR2 came out it would have some multi- second hangs. It that was one frame I'm not even sure it would show up as 0.1% low stats.
Maybe something that looks at frame time and will highlight frames that take 2× or 3× longer than the average to see if they are an issue at times other than loading screens.
Is there a way to measure 5% lows?
Only what matters its How often you will have these fps drops.
Less drops - Smoother feeling of gameplay.
Thanks for making this, anxiety over 0.1% lows has dissipated. 1% lows or above and frame pacing have increased for balance.
this is legit the 3rd video that i watched to learn what 1 and 0,1% low were the other 2 were too complicated explaining frame times and stuff
thx for explaining it simple
1% lows are the main thing I always look at in benchmarks. I never understood why the .1% lows were there, because it's literally a pointless factor. I would rather have a CPU that averages 100fps with 80fps 1%, than a 110fps CPU that gets 70fps 1%. You can easily pick up a CPU nowadays that run games flawlessly such as the 2000/3000 Ryzen and 9700/9900k.
But would you rather have averages 100 fps with 80fps 1% and 30 Fps 0.1% than 110fps 70fps 1% and 60 Fps 0.1%. The first example will result in 33 ms frametime every 10th second which is a stutter. A lack of memory can do that. I am not saying there is no case to be made for abolishing 0,1 (0,1 is noisy hard to replicate) but 0.1 is not a completely useless metric.
This was incredibly useful. I had a 4790k and I was looking at going to a 3700x or a 3900x and I was disappointed that the average frame rates were so similar but as you said its that 1% low where you feel the difference. games are much smoother; start up is much faster etc. Just mirroring what others have said you nailed it on the head.
I went from an i7-3770k to a Ryzen 7 3700x and the difference is massive, all the little stutters and slowdowns in games that I never managed to get rid of even by lowering the settings in the options menus are now gone. And I agree 1% low is just as important as the average framerate, but 0.1%low measures tend to be way too all over the place to be truly useful.
How about a frame time graph? I've seen some places give % of time with frame times less than X milliseconds.
Yeah. I have a build I will be working on later this year and the most important thing to me when it comes to CPU is how well it will handle frame pacing now and in the future. You are one of the few channels that I've seen that actually focuses on this aspect and it's greatly appreciated!
Tech Deals, you are absolutely right. My frame delivery consistency improved considerably going from a i7 2600 to a r7 3700x (as i expected it to). And yes, the upgrade was to improve the consistency between the lows and average frame rates taken as a whole and not my max frame rates. PS: I think you are on the right track in experimenting with how you benchmark and the way those results are presented to us. If anyone can figure out the best and most reliable way to achieve this, its you.
Same here going from 2500K (4.7Ghz) to 3800X (4.5Ghz). The frametimes are much more consistent so even 60fps feels nicer. What GPU are you using and did ytou change you GPU too. I had and still use GTX 1080. Honestly the platform upgrade pushed back my GPU upgrade considerably. No point in pairing 2500K with anything stronger if the frametimes are a mess.
@@Raivo_K Absolutely. That was a massive upgrade for you too. Great pairing with that 1080. Running a 1070ti at the moment and its like I got a whole new GPU with the new platform. But here is an interesting tidbit. I paired the i7 2600 with a 1050ti for a secondary rig and in newer more demanding games it bottlenecks the 1050ti (Detroit: Become Human) being the latest. Even in older games like Shadow of War. When I swapped the 1050ti into my R3700x rig the mins went up and turned it into a GPU bottleneck. I believe its the per core performance and the DDR4 3200mhz (tweaked timings) that help. The 1050ti is really low on bandwidth.
Anyway, what games are you playing and are showing the biggest improvements?
@@joeyvdm1 I don't play that many games at the moment. I was playing Mass Effect Andromeda multiplayer with friends. There i noticed the biggest improvement. Frostbyte engine loves 8 cores. Even when booting to single player my CPU was at 100% constantly for minutes before it stabilized and fps went to mid 50's.
Now i get faster loading, no waiting and fps is 90+ straight away. I believe all Frostbyte based games should see a nice per uplift. Also Quake Champions become much more fluid.
@@Raivo_K Absolutely agree. Multiplayer gaming and Frostbyte engines make great use of 8 cores (and i suspect more, to an extent, if you've got them). In single player gaming (Shadow of the tomb raider, Jedi fallen order, Red Dead 2, Detroit Become Human, etc) I have gotten major upgrade including load times (as you mentioned).
EDIT: All round system performance and snappiness is fantastic too. I do lots of video converting and I can encode with a gazillion tabs open while watching youtube and working in photoshop and it is still as responsive as if I was doing nothing.
Frame time variances as well as fps variances are the most important parameters.
They are achieved by calculating the derivative of those to parameters. That's what human vision is more sensitive about. I much prefer a constant 60fps experience then a 60 to 120fps, all over the place, one.
If the fps variance is slow and gradual the brain adapts to it easily but if it's changing by the second it is perceived as stuttering. That's very distracting and completely destroys immersion for me. That's the reason why I block my maximum FPS, with rivatuner, to a number that alows me to have the best and smoothest experience possible even if that number is well below the average fps achieved by my system.
That's a number between average and 1%fps and how close to 1% really depends on the game.
It's important for multithreaded processes like modern games. It shows to evaluate the expression of micro freezes.
I'd say if you do frametime chart (probably with nvidia frameview) it would be much easier to prove your points.
I think 0.1% low has its uses but the limitations are significant. It's best on built-in benchmarks and will often reveal things like microstutter. Consistently low 0.1% numbers are usually a legitimate issue and not just some random windows did something in the background. At the other end of the spectrum, an extended playthrough with respawns, saves, loads, etc will render it worthless. In the middle, on a stretch of playthrough that doesn't have those kinds of interruptions, bad 0.1% numbers are often indicative of some kind of real performance issue that merits further investigation.
Huge thanks and respect man......you have no idea how much stress you take off of my head....thanks a lot....
I don't know what to think.
My first reaction is to ask, would using standard deviations from the mean be better?
Instead of using the mean, would using the medium and standard deviations from the medium be better?
Your comment, it's better to have a consistent frame rate of 30 fps than have half your frames be 30 fps and have your frames be 90 fps, is very interesting.
That’s an option, the question is how to get that data easily.
Thx for this really interesting video! I totally agree with you. I also benchmark with Afterburner on my own machine. I just let it run in the background, while I'm playing. Sometimes the 0.1% is 0 or 1, although the game ran really smoothly for an hour. Maybe I just died several times or reloaded the save game. Steve form Hardware Unboxed never uses 0.1% lows. Maybe you know his channel. In my opinion 0.1% lows only make sense if the benchmark doesn't take any longer than 30s and is repeated several times to eliminate a potential outlier.
Hmm. How would you come up with an easy-ish test for frame consistency? Because I completely agree!
Great instructive video as always.
Always nice to have more people explaining what the benchmarks mean!
Thank you for being the only person I can find online who shows just how extreme 0.1% lows truly are.
I've now changed my benchmarking analysis to only include AVG and 1% lows in future.
Why do you think 3 and 5% lows might be better?
how do i keep the 1% fps stable? i noticed that when i play at 60 fps, 1% many times drops even to 30..making a fps drop..how do i keep it stable? or as high as possible?
Thanks TechDeals for tackling this issue. Average FPS is just that -- an average. And trying to sift frametime anomalies out of a subset of an average is froaght w/issues. Back in the day Scott Wasson at TechReport was championing frametime analysis, then PCPerspective experimented with some different graphing techniques, and GamersNexus continues with an occasional frametime plot. Sadly the more advanced analysis never became the norm.
I get on my rtx 2070 super and ryzen 5 3600 and 16gb ram 0.1% lows on paladins to 25 fps from 400 and i feel the stutter all the time how can i fix this temps are at max 60C
even tried to cap my fps still didnt work
Log fps for entire benchmark at set interval then plot?
I'm trying to understand frametime stuff better, so someone please correct me if I'm wrong. I think I disagree with the video on the subject of 0.1% framerate being useless. One of the points that the video makes is that 0.1% low is a bad measurement because it's such a small amount of frames it doesn't really tell you much about the whole experience. His example was that 0.1% of 60fps gameplay over 10 minutes is only 36 frames which is about a half second, which makes it sound like somewhere in the course of 10 minutes of play the game will freeze for half a second. That's not a big deal. But I understand it differently. If you're playing at 144fps, then the 0.1% low in 10 minutes of gameplay is about 86 frames. Rather than experience a half second stutter once in 10 minutes, that's 86 separate times that you're going to experience that 0.1% low. If the 0.1% low is bad enough, that's 86 times that you go to make a headshot and the game stutters and you miss the shot because of it. I might be completely misunderstanding this and if so then please let me know.
How much does capping the FPS/enabling Vsync affect frame consistency and stutters? All of your CPU tests are pushing the absolute maximum FPS and pinning the CPU at 100% (which naturally leaves few resources for system tasks), but many people play with Vsync or a cap and never get close to maxing out their CPU. It would be interesting if you could test frame time consistency between locked and unlocked frame rates to see if leaving some headroom could keep older CPUs going just a bit longer.
That’s a good idea...
This video is so good, i need this!
Thank you Tech Deals ❤
what would be a good 1% low statistic? my 6700xt always shows a numeric value of 2, I thought it was supposed to mean that sometimes (if my average fps is 120) then every once in a while my fps should go down all the way to 2? that shows extremely bad to me but I always have a smooth gaming experience. Then again I don't know much, any help please? is that a good statistic or not?
Gamers Nexus touched on this superbly in their 9600K review. The main thing I look for is frametimes, and if I see a spike on the frametime chart (which only occasionally show up in Afterburner's framerate graph), then my flags start going up.
ah got it. thats why i almost feel stuttery when gaming with my i5 2500 (non K) paired with GTX 1060 6G. maybe the time has come to upgrade.. 😅
i am on a i3 2100 with a 5700XT. imagine my pain :V
The time has come indeed. Sitting on a 3570K + 1060 and I'm almost always CPU limited.
Finally!! A video that explains 0.1% and 1% lows in a way I can wrap my head around. The key wording he used was "1% low means 99% of the time your FPS will be above this number". No other video (gamer nexus, RandomGaminginHD, bitwit etc....) about 1% and 0.1% lows took the time to elaborate on this one specific but important explanation. Great job!
Ive just upgraded from.an i5 4690 to ryzen 5 2600x keeping the same gtx 1060 and the experience is soo much smoother, yes im not really getting any more fps but it feels much nicer
GREAT Video!!! Between what I have learned from you and Digital Foundry I've been heavily focused on the 1% lows and .1% lows over Average FPS and learning to tune my graphics settings and cap frame rate to get a buttery smooth game play. I think a lot of us gamers (which I'm very much guilty of) are/were too focused on Average Fps numbers. This is why I upgraded from the 9700K to the 9900K because while the 9700K has the potential to OC to 5.2-5.3ghz the FPS increase does exceed what you would often get from a 9900K but what you don't get is the Hyper Threading to smooth out those 1% Lows for buttery smooth game play. For the record if I didn't own a 2080 Ti or plan to upgrade to a top tier card in the near future AMD is the answer because the cpu performance is a great fit and won't bottleneck the gpu but if you do own a 2080 Ti and play at 1440p you really should be pairing that gpu with an Intel chip to get the most of that gpu. Otherwise why buy the card? You would only gimp the performance to almost a near 2080 Super experience :/ I digress - JayzTwoCents ; in regards to the .1% lows I think you are probably right, its the 1% lows that seem to carry the most weight. Awesome channel you have hear man. I love the fact that you don't just review hardware and give the boring benchmark graph charts but you do lots of testing and demo what you are testing. This is how we learn, so good job man keep it up.
The only way 1% lows matter is if you calculate them via a circular buffer that constantly updates, so low numbers dont saturate the entire benchmark. You then calculate the amount of time certain 1% lows occurred during a set time period and possibly develop an average aswell. So that would imply a 1% low frame rate thats, on average, either closer to the highest FR or not.
so is not possible for the Ghost Recon example to have 10 fps in the 0.1%? that would be a couple of ugly skipped frames or even periodic stutters. Some skips and stutters while very annoying wouldn't appear in the 1%, unless they are a lot of them, but 1 ugly skip every minute would be awful. Just 4 runs show some variance but I don't think is too bad, just mean more runs are needed, for it to be useful, that "evidence", tells me not that 0.1% is useless but that 4 runs is too few. That data doesn't show randomness as you suggest, far from that. Of course it may be a better metric to shows frametime consistency, in that case great, but I don't think there is any evidence that 0.1% is useless.
Great video explaining the uselessness of the 0,1% low. May I suggest videos on combinations of CPU and Video cards that would pair together to give the "best" experience at 1080 and 1440 and 4K gaming. Or have you done this already?
I think your idea for measuring the frame pacing would work just fine 🙂
I dont know how good it would be, but maybe you could show us the median frame time, or the mode of all frames.
very important question,,,,, if your 1% low is lower than your monitor hz is that bad then? or is it not noticable at only 1% of the time out of 99% being higher than the monitor hz
Just to clarify, you can get better fps, or better timing, or faster frames by upgrading. I went from avg 50fps on an RX 590 with an FX-8320 on high settings in most triple A titles. I bought a 3800x and kept the same gpu and I now get 60-70+ fps avg with no lag or issues. So you can get better game performance by just updating the CPU. I'm not sure if you said what you said slightly off or if I misheard you. My apologies if I misheard you, but I hope this info helps. Also, it's not alot of performance gain, but it can be just enough to push you over on fps. The misses rocks an FX-8350 and I know there's gpu performance just sitting on the table, so I'm gonna try to get her on Ryzen for that same reason.
Edit: Rewatched video. I heard you wrong. And you covered all I said above. Thanks for the video!
Great explanation . Looking forward to future benchmarks
One possible way to test is to cap the max fps at say 60 fps. Then the average fps becomes meaningful as extremely high fps wont be able to cancel very low fps and hence will accentuate performance dips. The closer they can get to the cap (60 fps), the better the card is.
Tech, sorry to comment before watching the video, but can you please rename the video with the word "very" in front of "useful" just for the sake of importance.
Be assured that I'll watch the video when I'll finish work, keep up the good stuff for us, you're awesome!
Showing frame times sounds like it would a useful addition to the regular average and 1% lows.
This video sums up why Tech Deals is a great channel. Well, sums up one part of it.
1% lows are why I get frequently upset with my 2080ti not even being able to hold any amount of fps steady in any of the game that I play.
OMG someone finally broke this down for me 💯 thanks 👍🏾
Thank you for the great vid. I like the idea with measuring the responsivenes of a cpu.
What would be great is if you could get the time intervals between frames.
Then you could calculate the variance and plot the data
Thanks! Very interesting! You mentioned 3% low etc. Wonder what info afterburner dumps from the benchmark and if you can calculate it with Excel using the perceintile function?
I love your work and how you truly explain everything. Before I ever make a purchase I always go over your videos on it. And I have been meaning to tell you, love the new intro. Keep up the good work.
I'm glad you are looking at the commonly used benchmarking system today and identifying it's flaws. I agree to omit .1% since it's not significant to the experienced gameplay. I will suggest a benchmarking system which makes a graph of fps versus percent to show how 'often' you can expect a certain fps. A narrow bell-shape will be better, a flat dome-shape will be worse. With this, one can see how much the fps varies, how often the fps are under accepted level, and get a better view of frame-pacing.
Good job, Tech. So, when shopping for a non-existent graphics card, start with "Avg FPS", focus on 1% Low, and ignore 0.1% Low. Got it, Thanx.
Hi, I need some advice regarding of gpu, I was planing to get new gpu, and at first I wanted to buy rtx 2070 super, but after recent events, I started to consider buying rtx 2060 super, the price for 2060s is ~480$ and 2070s ~630$, so here my question is it worth it to pay extra 150$ for 2070s, or 2060s will be enough for 1080p maybe some 1440p for the next few(5) years ?.
The human eye can't even see 1% or .1% .
The human eye can only see 24% per second.
Can I do anything about my 180 FPS AVG when 65 FPS 1% lows in Call of Duty Warzone? Ryzen 5 5600x + 5700 XT + X570 + 32GB 3600Mhz CL16
it's like riding as a passenger in a moving car
situation 1 passenger = a driver drives at 120mph and sudden brake to 40mph in a second and goes up again to 120mph next second
situation 2 passenger = a driver drives at 60mph and sudden brakes to 40mph in a second and then goes up to 60mph again next second
i would choose "situation 2" passenger experience rather than "situation 1"
Heading in the right direction but when u measure frame times you end up in same spot with %lows
for me it shows 0 and 1 fps.., my game runs at 50 - 60 fps, no idea
I agree with you, Tech. When looking for at-a-glance performance indicators for a given set of hardware, I always assign 50% of the score to the average frame rate and 50% to the 1% Low. Both of those numbers are very useful.
Oh and adding to my other comment:
The apparent difference in accuracy between 0.1% and 1% is merely a function of how long your benchmark is. It just so happens that you (and most people) tend to run benchmarks for long enough that 1% of the sample is still meaningful but short enough that 0.1% picks up "noise". If you ran the same benchmarks for 10 times as long your 0.1% results would become as meaningful as your 1% results are now. In turn, if you ran shorter benchmarks you might have to give up the 1% category.
Assuming that the "noise" really comes from outside the games it should be constant, at least on the same PC with the same programs running, so we could calculate a precision cutoff depending on the length of each benchmark. So for longer benchmarks we could go all the way to 0.5% for instance, and for something like the Far Cry 5 benchmark we might stick with 3% or whatever.
Should I cap my frame rate to the 1% lows? or around the average framerate for the smoothest gameplay? Specifically Esports games. For example on Apex I average around 80fps and have 1% lows of about 50fps.
It will result in input delay if you cap your fps, keep it uncapped my recommendation
Also appreciate this was a more bite-sized video! I love your videos but prefer 10 min to 30 min videos as like to watch youtube inbetween doing other things!
I mostly agree with your statements. 1% low is definitely enough when 0.1% is above a number where you notice lags and jumps. When the 0.1% value is 30 or even 50, then you'll never ever notice the moment when it happened. But let's say you measured a value of 10, I bet, you could tell me the moment when it was measured given the fact, that some things were moving in the game.
you nailed it. awesome!
If you were a teacher in school all studs would come first...cause you explain so well....
Excellent explanation! Thanks!
Great information. Thanks!
Can you do a list of best budget options for AAA gaming in the different resolutions for the next three years and or 5 years?
After finally starting to do all my benchmarks to include 0.1%, I immediately I realized how stupid it was to rely on this specific metric as a critical performance indicator. As a person benchmarking a system, it truly means nothing to another user who will have a different program load on their CPU. The ONLY WAY to selectively benchmark the minuscule window of “good” 0.1% for the sake of being able to show impressive numbers is to play in such a way that does not at all represent a typical usage scenario. I’m with you on this one. Great breakdown.
I don't like any of the current options. I'd like to know what a 10% low metric would look like.
Why not create a probability density function and use standard deviation?
How about using standard deviation as a measure. I would believe 60FPS with a std deviation of 5 fps.
I was going through the comments to see if someone would mention it. Or SD expressed as a percentage of the average FPS.
I totally agree with you a static frame rate or a frame rate with low variance is better than when it jumps all over the place. Especially with synthetic benchmarks, sometimes the frame rate may ideal around 25 fps than jump to 75 or more for a few seconds just to return to 25 fps. I wonder how many people realize that is just how the system behaves depending on how demanding the game.
Input lag will be better. Knew I wasn't crazy. My i5 6500 with a gtx 1070 completely ruins my gaming experience. Input lag is mad idc what fps i get. Is all of this linked to cpu usage in any way?
HI! How did that 1% low to be average? Tnx :)
This was a very informative crack at a complex problem. Average fps rates are easy to understand but the truth of the gaming experience is more nuanced. Unfortunately, people have increasingly short attention spans and less and less time to consider complexities. Sound bites and summaries only, please! Sadly, this is the equivalent of information fast food.
.1% lows basically tell us if we can expect to a game to have severe hangs and snags. Its significance doesn't come from what it tells us about a framerate but rather what it tells us about a game's stability. It's certainly less useful than the other statistics considering that improving it is often much less straightforward than the others if it can improved at all, but it does serve a purpose.
Thanks that was very helpful.
How hard is it to get the standard deviation? Give me the average to show general performance and then something like the breadth of the 99th percentile. For example, one rig might produce 60 fps with 99% of the frames within +/- 15 fps while another gave 60 fps with 99% within +/- 20 fps. The gap between 1% lows and highs would indicate the smoothness, lower being better.
sometimes is useful talk about the gaming experience and not only numbers. I have a 9600k and his fps stats are high but it feels a little trash regarding the gaming experience. Waiting a motherboard for a 3700x and check if the gaming experience will be better. thanks for that video
So, I need to upgrade my i7-2600? I thought I have at least year or two.
@ZhekaTrololo
It depends bud. If you are happy with the performance (and by your statement it sounds like it) then wait until it isn't good enough for you anymore. If you are asking the question because the thought hadn't crossed your mind before (because it is all good for you right now) then no. Wait until you aren't satisfied anymore. PS: I wasn't satisfied anymore with my performance in newer games, so I upgraded.
@@joeyvdm1 it depends also on possibility to sell it. Maybe later it will be more difficult.
@@joeyvdm1 Yep. I had 2500K and while in older single player games it was totally fine it struggled massively in newer AAA and especially Multiplayer heavy games. So it also depends a lot on what you're playing and what GPU you are using. Im using a GTX 1080 so i always knew my 2500K was holding it back somewhat. Even at 1440p high settings but upgrading to Ryzen 7 3800X last year gained me about 25% more performance despite still using the same card. Especially multiplayer games went from stuttering or occasional hitches to buttery smooth.
So with the old CPU i might have had 60fps but the frametimes were all over the place. 2600 has HT but i feel like even that is starting to show it's age. Newer CPU's don't add just cores. The IPC and thus single core performance also goes up nicely.
change to high settings . thats it . utlra settings arent meant to b played
@@ZhekaTrololo No one knows your situation better than you bud. If you believe you have a better chance of selling your current rig to help finance costs for an upgrade, that is an incentive to do so.
Okay cool. Let me see if I can help you.
What is your current rig (GPU,RAM) and what would you like to upgrade to or have you no clue yet? Are you looking for suggestions?
I had a i5 9600k 5 ghz, the Witcher 3, which are not the newest game by far, sometimes made some strange stutters, just for a second, I hated when it happened and It was very annoying. And I started to monitoring my CPU usage, GPU usage, FPS, temps etc. and that moment, when those stutters happened the 6 core CPU did run up to all core 100%, but most of the case the CPU worked on a high precentage. I have Steam, Epic Games app, Origin app etc. running in the background all the time, also CAM software for my Kraken x62 AiO, sometimes I forgot close the browser. So I did some research on Reddit and they said to me go with a higher core count CPU, because that could be the only problem, I have an RTX 2080 + 32GB Corsair Vengance 3ghz CL14 memory, so I sold my Intel mobo and CPU and bought a Ryzen 9 3900x, becaues I considered to start making youtube videos again. Anyway all of these hiccups are gone, I have a much smoother experience, even though the avarage FPS dropped a 5-8 fps, but who cares, the 1% lows are elevated from 10 fps to 80 fps, that's something and I don't have to bother anymore the background apps, they can run without problems!
I’m getting 2 FPS for 1% lows. Is this a problem!
msi gtx 1660 super gaming x or the new rtx 2060 KO from evga?
I'd love to see a new type of graph for frametimes. For game benchmarks where you can get a count of frames and a log of frame times, quantize them into millisecond "buckets" -- e.g. take number of frames that took over 500ms to render, then 250ms, 125ms, 66ms, 33, 16, 8, and under 4ms.
Plot frame total on the X axis, and plot frametime bucket on the Y axis with slow frames at the bottom, fast frames at the top. Connect the plot points to create a curve. I'd call this a "FrameTime Quantization" or FTQ for short. "Good" results would have a sharp peak as high and to the right of the chart as possible. "Bad" results would have multiple peaks. The tallest peak would equate to a "median FPS" - e.g. a sharp peak at 16ms would be a "solid 60 FPS".
What I would like to see is the end of "fps". I believe the discourse should always be about frame times instead of frame "rates". Why? Because 100 fps can feel great (Battlefield 2) and they can feel awful (CSGO) so we learn nothing from that number. We need metrics that get to the heart of what we want to measure and that's how responsive games feel.
To me the "1%" and ".1%" terminology obfuscates even further what really matters than the standard term "framerate" already did. When it comes to responsiveness we care about the attributes of EACH INDIVIDUAL FRAME, such as:
Latency - how old is it, how long does it stay on screen?
Synchronization with the monitor - is each frame complete or are we getting partial, torn frames?
As far as I'm concerned ONE bad frame in 10 minutes ruins everything. If a game runs at 1000 fps for 10 minutes, except once a single frame stays on my screen for 50 ms the game has a performance problem. 1000 fps and bad performance at the same time! So why talk about fps? What matters is the responsiveness I can RELY on. If you go climbing with a rope that holds your weight 99.9% of the way up a cliff the final rating for the rope is 0/10, useless. You fall off the cliff and die.
The 1% lows in frame rate or peaks in frame time get closer to the heart of the matter but really even they are too generous, that's where the desire for 0.1% numbers comes from. If the game fails to respond reliably at any one point then the game is overall unresponsive, no matter how many frames it pumps out before or after. True, reliable responsiveness is what we need to measure, and we need to figure out how to dismiss "noise" caused by Windows quirks or driver bugs or whatever, without losing sight of the ultimate goal. The ultimate goal is not and never has been maximizing the number of frames. When you look at how terribly inconsistent modern games are I also suspect that game developers have lost sight of the goal and are optimizing in ways that don't actually lead to responsiveness.
I always ignore .1% lows and disregard any benchmark channel who use that just to trash games they don't like, even if the average and 1% are fine and that .1 may just be a momentary stutter due to whatever reason.
well 0.1% might just be a buzzword created by the tech industry journalist to entice the viewing public when all it really matter is the credible reporter saying "this game has a superb frame pacing and frame delivery on xyz hardware meaning youll have a great and very smooth gaming experience." in my experience Ubisoft is the culprit in this hulabaloo when they released watchdogs 1 requiring more than 2 threads/cores to run smoothly as having less means 0.1% or even 1% low of 0fps meaning the game pauses from time to time to catchup and it totally breaks the immersion experience. sadly this is now rampant on most games. i also remember back then in gaming we never cared for slowdowns framedips etc, we just look at it like it is part of the maximum experience the game devs are trying to offer, ridge racer 1 on ps1 was awesome even if its just 320x240/30fps and dips occasionally while the arcade is rock solid 640x480/60fps with way better textures. we have been spoiled! imho we need a new display monitor technology that combines the best features of flat panels and crt's, as most of pc gaming buyers decisions revolved around that pesky hard to drive flat panel monitor (yes monitor dictates what gpu youll buy then the rest follows).
How about 1% highs to see how much frame time varies