What Are 1% & 0.1% Lows?
HTML-код
- Опубликовано: 7 сен 2024
- This video explains our measurement and presentation of framerates in GPU & game benchmarks. We define what "1% LOW" and "0.1% LOW" values mean in relation to frametimes, average FPS, and game fluidity.
Recommended: Interview with Scott Wasson - • Scott Wasson on HDR Di...
Recommended: Inside the Second - techreport.com/...
Like our content? Please consider becoming our Patron to support us: / gamersnexus
Host: Steve Burke
Video: Andrew Coleman
** Please like, comment, and subscribe for more! **
Follow us in these locations for more gaming and hardware updates:
t: / gamersnexus
f: / gamersnexus
w: www.gamersnexus...
Links to Amazon and Newegg are typically monetized on our channel (affiliate links) and may return a commission of sales to us from the retailer. This is unrelated to the product manufacturer. Any advertisements or sponsorships are disclosed within the video ("this video is brought to you by") and above the fold in the description. We do not ever produce paid content or "sponsored content" (meaning that the content is our idea and is not funded externally aside from whatever ad placement is in the beginning) and we do not ever charge manufacturers for coverage.
Very informative and useful. I always wondered what 1% low and .1% low meant.
How can you not know that Rick?
Because he is dufus Rick
uyaratful Oh i see.From an alternate univerise.
I hope you're kidding
CPU or PCH on the GPUs with undervolt and overclocked applied with XMP OC on the RAM and on the motherboard BIOS settings or not?
You basically humiliate larger channels with your effort, methodology, techniques and of course, the hair.
N1 malaka
That's tech Jesus to you. He's friends with Linus, JayZ, and the other techtubers out there, so they aren't exactly competing against each other.
@@stevethea5250 haaaaaaaaaaaaaaaaaaaahahahahahahahaha
@@triadwarfare yeah got me scratching my head, not my butt.
It is clearly he could write an essay form easily, while some others just try to be first on anything and can't accept the fact there are other people better than him (as a life experiences, you would meet those kind of guys frequently.).
As a engineer and statistician I would like to commend you on providing some helpful insight into basic statistics for your audience . . . I've been a long time viewer because clearly, you have the background and expertise to correctly interpet & present your data while so many others on RUclips (and their audiences I suspect) simply market often unfounded opinions. Good job! -rk
No gray hair steve, younger steve. Its been quite a ride.
And NOW, i KNOW.
Thx Steve
I used to look only at average fps to decide which video card has a better bang for the buck. Now I realize that frametimes represented on your channel by 1% and 0.1% lows are as important if not more important. Thanks for the great explanation and taking the effort for going more in depth about this topic.
Hey man, cool profile pic :)
@@TheSpiritof76you too
thank you Steve from the past
This type of insight and review will hold manufacturers and developers more accountable. I love when something like this occurs which forces various involved parties to be more honest.
he just loves to quietly show off that 1060 doesn't he
quick, go tell AdoredTV, its all a cabal to end AMD
I'm on mobile I can't like this comment 😥😥😥
+DemonSage I have the same problem. Damn mobile app, how hard can it be to code a proper comment section?
4 years later that gpu is like 10% of a 3090
What a youtube comment
It's closing in on the end of 2019 and I'm still having to show this video to help explain things to folks.
The year is 2024 and I'm getting informed.
Thank you for this description on what it means. I was worried about the video being 10 minutes long, but it's definitely not something you can properly explain in two, and this method of measuring performance makes so much more sense. I'm glad you do it this way, and now I know what to look out for.
I love how much though and foresight you guys put into everything from your test methodology to your articles & videos. I recently started supporting you on Patreon and am glad to have done so.
I've gained an appreciation for this metric since playing GTA V on SLI 970s. Framerate hovers at 80 most of the time but even dips to 60fps feel terrible as the frame times are all over the shop during the dips. Really appreciate you going to the extra effort to report these metrics!
One of the most informative videos on GPU benchmark. GN, sir you did good job here. Many ( majority ) tech reviewiers should be linked to this video.
Thank you for pouring your heart and soul into your work Steve. Although I do not always agree with your conclusion, and those are rare instances, I will say that your testing methodology is solid and by taking the frametime metric and incorporating it in this way gives a much better actually perceived value. I have had cards run 'faster' than others that play like crap and multi gpu setups before the whole frametime issue was uncovered that were blistering fast but horribly unplayable. Great video and thanks for the hard work to help consumers make better decisions.
This is great measuring idea! Great that you adapted it.
But is there any way to minimize effects of 1/0.1% lows? Can you do some shenanigans to lower your framerates and make frames more consistent?
Great question! I was meaning to take this a bit further and ask if AMD's frame target utility actually helps in that direction, i.e. could you trade peak performance for frame time consistency ? If so does AMD's solution achieve that or is it not working that well ?
One thing is *not* to run SLI as that tends to induce a lot of frame time variance. Another source of variance would be if you were pushing your GPUs RAM and it periodically ran out and had to swap memory out to the system - that'd take extra time and cause a frame time spike - so check GPU RAM usage and reduce graphical settings to bring it down a bit.
If it's caused by low VRAM you could try lowering your texture quality or resolution.
I know this is an old post but what works for me in games with big FPS swings where I normally have some headsroom is to use Rivatuner to cap my fps at about 62-65 fps depending on the game and disable all forms of vsync.. This keeps you just above 60 but also keep the GPU or CPU from hitting a hard stop/stutter so it still has the grunt in reserves to cope with whatever would normally make the stutter.
That moment where everything makes perfect sense xD.. Subbed
This is by far the most ground breaking video I've seen in a long time. Thank you!
I always suspected that the "Minimum FPS" value doesn't mean much. Excellent video! Subscribed.
Young Steve is so sweet
(not that the current version is not, but you know what I mean)
This is a very rigorous and reliable metric of performance. Channels like 'Science Studio' give the min/average/max data. Ironically your methodology is far more scientific.
GUYS SO WHAT'S BETTER? HIGHER NUMBER OR LOWER FOR 0.1 AND 1.0%????
@@stevethea5250 higher is better
@@poizonedcookie2669 thx
Two questions...
So now that we know what they are, how can we make measures to improve 1% and 0.1% lows?
Are these more so cpu constraints causing the dips to begin with or gpu constraints?
i always wondered about this when playing games, but never found the words for it other than "lag" or "stutter" thanks for doing all of this.
It's so dope Steve. You're such a badass. Seeing how far you guys have come is incredible. Favorite Channel For sure.
Ok cool makes sense, I like it when you used the 0.1% lows cus then when you benchmark a game (for example battle field 1) when I see a really low 0.1% score I can expect that game to stutter, and it usually does, as I said ex : BF1, DX12 as you saw in your testing had big stutters and it did indeed stutter really bad with my R7 370.
Finally I understand it too.
And it's nice to see, that steve didn't change a bit since then
Another really good video - these guys make great, informative and vendor neutral content.
I find the rise of the tomb raider benchmark a classic example of why average frame rates are not what they are cracked up to be. Unfortunately it doesn't give 1% and 0.1% lows but the lows are so common that to hit a frame rate where I rarely dropped under 60fps, I needed to drop the settings down substantially on my 390x.
Apart from bragging rights, what use are fps averages anyway?
It's basically the cumulative distribution function or CDF. For the layman, it's easier to say (in human terms) that 99% or 99.9% of the time it is XXX FPS, rather than 1% low or 0.1% low respectively
6 years ago, he was so YOUNG!!!
When there is a big disparity between 1% and 0.1% low in a particular game, what is supposed to be done? Turn off some effects until stability is reached?
get a new card
I've heard tuning your RAM can help with that, but have no personal experience to back it up.
I've always like that you showed 1%/0.1% averages instead of minimums. Max frames per second is a completely useless metric that other reviewers show, and I've always been hesitant to trust minimums for fear of outliers
Wow, I'm watching this video wondering, "That's pretty old tech there"... lol, the tech is old, but the info is fresh as always!
You have OPENED MINE EYES!
How can you have such a small group of subscribers. I really like you scientific approach. I just started watching your videos, but they make so much more sense than many of the "big" channels......not naming any, but you know who I mean.
Great channel, keep up the good work
I've been wondering what those stats were for months now. Thanks guys!
I really like the way you present the information, it's what attracted me to your channel. I wish other reviewers would adopt the same metric.
Great video, and a very eloquent explanation.
This helps me feel a bit better about going for the 8GB RX 480 rather than the 4GB.
holy , look so young here
He also sounds really bored lol.
Great video, and useful discussion moving the metrics to where they need to go. @Sergei, agree, but - small steps. We're almost halfway there if we get people to start understanding and demanding it from other review sites...
2021.....just found this. Great explanation!!!!!
Excellent methodology. I was sick of whole "average fps" thing as it shows unreliable data. Instant subscription because of this work. Thank you.
Average fps does not mean you will have a stable gaming fps of that number. I would prefer stable 40 fps to 60 fps if i will have spikes and random fluctuations.
Nice one Steve, as usual the best info anywhere on the planet.
On games I worked on in the past we'd do memory defragmentation or in newer games garbage collection during a scene transition when the screen is black. It would be unnoticeable to the player but would look really bad on the profiler if you were looking at the worst (longest) frame.
6 years ago, wow. You’re like baby Jesus - perfect time of year for a nativity scene.
Can you please explain in the year 2019 anew what 0.1 and 1% lows exactly means, because I have a hard time to understand it. I know what stuttering is, but what is a good value for 0.1 and 1% low?
I am SO GLAD I came across this video and explanation. Thank you!
I liked when you put the frame times of the GPUs in the DOOM Vulkan comparisons
This is why I hold the GN benchmarks higher than everyone else's. Best gaming-related tech channel on RUclips, and that's not up for discussion.
I'm having some strange situation where "minimum fps" has higher results compare to "1% low". how is that possible??
Have been watching some of your content since the rx 480 hybrid videos. I like how you go in depth with the testing and explaining technical details. And the janky modding stuff is fun to watch, so I subscribed! Keep up the good work.
OMG, such young and such camera shy TechJesus. :D
i really dont understand these things but i sub because you are so cool
It's really informative. Thank you! I went through the article as well. It's great.
Hey man I found you really knowledgeable and easy to understand.
Also if anybody (my friends you know who you are) hate on u for ur features hair and pimples don’t listen to them. In recent times youtubers have been getting a ton of flack for their appearance.
subbed! wow what a great explanation of a complicated topic. I now have a better understanding of how to look at component benchmarks critically and take things with a grain of salt. Hope more channels start looking into these metrics deeper and giving people a more full picture of overall component performance and not just one limited indicator.
2021 - Thank you for this video
For that 4gbs vs 8gbs testing though, you could always drop the textures or something to use less VRAM. More just a note for the end user, Game X at Ultra Settings on the fury X with textures set to medium/high. Probably not worth the time to include and test that though.
the 4gb vs 8gb test is a little crazy, if you use less than 4gb its basically going to be a draw if you use more than 4gb its going to be a win for the 8gb card. Kind of doesn't need to be tested. I think of it as being like asking what can hold 3L of water better a 4L jug or an 8L jug, they are both equally capable. If you asked the question which can hold 6L water tho there is obviously only 1 winner. Note in this 4gb vs 8gb test it wasn't a clock for clock false test it was a true to life out the box performance test as the 4gb models have 7ghz memory and the 8gb have 8ghz memory, G.N. tested it the right way and did find the benefits of the 8gb card.
Do you use Fraps or FCAT for measuring frame times? :) Although I really have no idea how FCAT actually works..
This explained a lot! Thanks for uploading this.
1% lows is real, I had a RTX 4070 Ti @ 1440p running mw2, Ultra, Extreme settings and though i was getting 160fps with DLSS on, Every so often i would get this hitching which is from lower bandwidth = 504GB, Memory bus = 192bit on the 4070 Ti which tells me the 1% lows where iffy, Pretty sure there was a bit of a bottleneck there
So that said you cant always go by AVG or max framerate. Granted the gpu held its own for the most part, Seeing a 4070 Ti push that amount of fps was sick but the 1% lows (Hitching) was kinda meh, Now to some it may not be fully as bothersome as it wasn't like a normal stutter but it felt almost like micro stutter as if the gpu missed a step that's the best way i can put it. RTX 4080 with the same exact settings would of eliminate that im sure
With that said, Seeing that sorta gives me a indication of what could approach in the near future with the 4070 Ti, If MW2 is doing that now, What would MW3 bring or any other game for that matter.
Can you do a video on the question whether lowering max frames in favor of higher 1% and .1% lows works and with wich limiting processes?
That would vary game-to-game. General advice would be just reduce graphics settings to reduce the load on the RAM and thus the chance of spikes.
johnnypopstar yeah sure, even I can think of lowering settings to gain fps and increase lows.
I'm still interested in what i asked^^
Thank you for this very informative video... I'm just getting back into computers and maybe gaming so this is a great help.
I wish we could see the raw data of the 0.1% frametimes. That’s like 6-15 frametimes to report depending on the avg framerate and length of the benchmark(?) There’s a case to be made in knowing these minimums because even an average of the lowest 6-15 frames can be misleading. For example, if you get 3-4 significant frame drops (let’s say double the frametime avg of the entire dataset) over a two minute benchmark and the rest of those 0.1% ft’s are close to average, then the 0.1% low could still look pretty good even though most of us wouldn’t like 3-4 stutters every 2 minutes. Maybe most stutters don’t work this way so it’s not worth showing but I’d like to know what hardware has the least minimums since sutters are a real pet peeve of mine
I think you are bettering the process of informing people about the Products. It almost comes across like the FPS avg / min / max information we are provided (elsewhere) is what you might describe as "filtered." I had a similar thought process regarding the power consumption of the GPU because it acts similarly to FPS with broad range of spikes (both up and down) possible. Do you think a similar metric can be used to better qualify the power data because what I believe what we are seeing is a similar filter applied as to the FPS measurements we're accustomed to seeing?
Stevie's just a baby!
after 6 years i getting this in my youtube video had to watch it lol
Great video, very interesting and needed ! Thanks :)
amazing video
Perfectly explained :) and i totally agree with this method of testing rather than conventional. I was having this issue and was thinking that why game is lingering on 60fps with that much frame rate latency.
Great video, thanks for explaining this.
i hope the other reviewers follow suit to your format of 1% and 0.1% lows.
as a new subscriber this video was great. Thanks a bunch
Amazing video, this is enough for me to seek council from GN first in the future. Perfectly explained to my knowledge level, without over or under simplifying
Thank you for the explanation.....Now that I understand, I prefer this method.
One of the most helpfull videos when it comes to understanding FPS but almost no one seems to care if you look at the views. Funny world.
thanks for explaining this in greater detail!
thx for the knowledge friends
Thanks for this explanation!
Now I understand that 1% and 0.1% are important. Thanks for the explanation! ♥
excellent explanation, subbed!
guys u r one of the best tech channels!!!
Super useful video, will be doing this ourselves in the future 😃
@thegamrone you didn’t show the 1% and 0.1% lows which shows the stabilize fps
I've just come to watch this thrice. I feel dirty coz i just watched Bit-of-a-twit's copy!!
Is the 0.1 and 1 % low measured by msi afterburner representative or useless?
So what's the best thing we can do if we are experiencing the bad drops from .1% lows? Change settings or hardware?
Depends.
i thought it was quite obvious what it was, oh well, nice to see an full on explanation on it, information doesnt kill
If I've understood you correctly then all using 1% and 0.1% values does is give you a better estimate of the "minimum"-ish framerates, you're still not considering frame timings. For example an FPS-Time chart that looked like /\/\/\/\/ and /|/|/|/|/|/ could easily have the same average, minimum, maximum, 1% and 0.1% values, but clearly the former would provide a much better UX due to the changes in framerate being much more gradual. Why not actually use the frame variance figures instead?
From a statistical point the 1/.1% low presentation doesn't really add too much. Far better than just a min fps but still incomplete. You should present it as an avg fps using standard deviation and then demonstrate outliers like lows outside that range, percentage of frame points at the low end and avg time the drops are greater than the 8ms range you mentioned. Overall you can create ranges of data points which would be more descriptive of general real life gameplay.
Well explained
how's the rx480 endurance test going sir?
I was wondering the same. By the way, great video as always Steve!
Maybe a simple standard deviation would be useful?
Thanks for the explanation :-)
Once, people used to compare CPUs solely by their clock. I hope we stop throwing around FPS (or worse, sometimes avg fps or even max fps) as a good measure.
I don't think an fps time plot is necessary, but some values like these lows are probably enough for me.
Looking at this video in 2020, and it was a great explanation of what these numbers mean. But I also have to wonder: what causes these 1% lows? Is it CPU-related? Does a CPU with better IPC make these numbers better? I am asking because I genuinely perceive these dips very frequently in Fortnite, usually in the lobby area, but also during gliding down from the bus to the ground and even on occasion in certain areas of the map. I am on a Ryzen 2600 CPU and have a GTX 1080. Would a Ryzen 3600 or even the new 5600X help with these dips?
Hello from 2021 i have r5 3600 and a 1080ti i have these dips in all games would love to know how to fix them. Some say go intel
@@_sdorts_718 If they are like micro blips, that could be a frame time issue that I've seen in the past. It's where, yes you're getting a consistent 60fps output, but the frame time moves around between 12 and 17 ms, where it should be a consrtant 16.6ms. That is with me assuming you've got V-Sync on to avoid screen tearing. If not, then ignore the next suggestion. So, if you do have v-syn on, try to use MSI After burner, then click on the tray icon for the Rivatuner, and put a 60 for frames per second instead of 0. IN the game, you may now set the FPS limiter to unlimited (you still get only 60fps since Rivatuner is limiting the frames). And your frame times should have stabilized to 16.6 ms.
Other reasons for dips if they are not of this nature could be sudden loading spikes from the HDD (if you still run from one of those), Windows doing some background task that momentarily takes away from processing the game. And yes, Intel CPU tend to have better IPC numbers that help with draw calls to the GPU.
I have since upgraded to a Ryzen 7 5800x, and I couldn't be happier with the CPU (aside from its crazy heat which can be worked around by undervolting the CPU). I also now have a RTX 3090 FE, and I still sometimes get blips in games, especially Fortnite in waiting area as players load in. I chalk it up to the game being online 100%.
Hope my reply helps you a bit.
EDIT: You may also want to checkk in nVidia settings for the Power setting. If it is set to " Normal" (default), your GPU might run at lower clock to conserve power, but clocks up when it's needed, which that split second can also result in a small blip. I also would try going into the Windows power settings, and if you are running Power saver, you definitely wan t to change that to either Balanced or even High Performance. So: for maximum imopact, I would go High Performance for Windows power plan and with Prefer Performance for nVidia power settings. Also, make sure your case gets good air as thermal throttling on the GPU can also lead to blips.
@@bellotriggerfish yeah ive done that with rtss. I can get all games smooth with gsync 135fps 7.2ms 1080p low setting and i get these spikes to 10ms 15ms 20ms to highest ive seen is 50ms spike. Gpu usage drops cpu usage drops game stutters for a split second. Setting ram to use 2T kinda made these spikes a bit softer but they are still present. Got m.2 and ive noticed that these frame time spikes come when i move in a new area of a map. I kinda think i need a better cpu i got r5 3600 with 1080ti gskil 3200mzh cl14 ram . Ive been trying to fix these frametime spikes for months.it f***s with my aim in fps games.
@@_sdorts_718 Yeah, so CPU does become more important towards higher frame rates. I do not have any monitor, only a TV. So I run everything at 60fps. So yeah, you're potentially beter off with an Intel CPU for their high IPC count. Those, again, can affect draw calls quite a bit. There are videos that compare GTA 5 AMD vs Intel unlocked frames, and Intel clearly takes the cake. As you know, Intel is a bit more pricey, although AMD has also gotten more expensive. Plus, you will most surely need a new motherboard, whereas your current board for the R5 3600 will most likely be able to made compatible with the new Zen 3 chips. I have an Asus Prime X470 and was able to upgrade to the new chip with a Firmware update. Just food for thought.
@@bellotriggerfish yeah i got b450m mortar max. Good board. I dont think i will upgrade anytime soon to a intel. Ive built this pc early this year. Dont have to mutch money to throw around. Ill just try to ignore those spikes. And thx for clearing things out for me a bit was thinking that i have a hardware isue.
This was super informative.
Here from hammer on box's poll, I'm glad 70% of people were right. This makes much more sense than 99th percentile low to me
thanks, was really easy to understand from those , sadly i have exactly those things now in watch dogs 2 1.07 with 2x R9 390x + I7 6700K
Hi GN! loved the video! as always, immaculate work - as an engineer i learn a LOT from you (your report quality passes mine x10 times)
edit - had stupid question, some napkin math made me chill again :D
still, part of it is, do 1% highs matter?
Finally, I've been wondering this a lot
thank you for the info! :)