Question about 4x-16x Anisotropic Texture Filtering and Performance

Discussion in 'Videocards - NVIDIA GeForce' started by BlindBison, Jul 7, 2019.

  1. BlindBison

    BlindBison Ancient Guru

    Messages:
    2,404
    Likes Received:
    1,128
    GPU:
    RTX 3070
    I've read about this in the past and most people it seems (including Alex from Digital Foundry) recommend simply setting Anisotropic Texture Filtering to "max" (16x) globally in the control panel and forgetting about it more or less since the performance hit is arguably negligible.

    My question then is, is this actually true?

    I've read benchmarks from TweakTown for Overwatch for example (and some other modern games) where increasing this setting from 1x to 16x does have a discernible performance impact on modern cards despite what the general perception seems to be.

    Now, of course this penalty appears relatively small in these tests -- 1 or 2 fps basically between 4x and 16x for example, but I might argue that can be a meaningful difference in some cases.

    Perhaps notably, I recall on last gen consoles something like 4x ATF was generally used as the "sweet spot" for this setting in many cases.

    For myself, I have a very hard time telling the difference between an image at 8x AF vs 16x AF so I wanted to ask about what the recommended value is for this on a forum where I expect people actually know about this sort of thing.

    If the performance impact is truly negligible and 16x is the preferred default for this across the board, I'll start doing it, but I wanted to verify this first since 16x seems like major overkill unless I'm missing something.

    Thanks for your time, I appreciate it.
     
  2. -Tj-

    -Tj- Ancient Guru

    Messages:
    18,095
    Likes Received:
    2,601
    GPU:
    3080TI iChill Black
    Some games had issues with 16xAF.

    ID5 engine in general

    For example BatmanAK ran smoother at 8x, no hickups. And few more I can't remember atm
     
  3. jbscotchman

    jbscotchman Guest

    Messages:
    5,871
    Likes Received:
    4,765
    GPU:
    MSI 1660 Ti Ventus
    I've been running 16x AF for years now, no performance hit at all.
     
  4. Apparatus

    Apparatus Master Guru

    Messages:
    429
    Likes Received:
    73
    GPU:
    3080 ti Vision OC
    Globally I set it to Application controlled and the Texture filtering quality to High quality.

    I check the optical result on game basis and then I use nvcpl or game options.

    Always 16X though, cause the performance hit is insignificant in my opinion.
     

  5. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    4,387
    Likes Received:
    627
    GPU:
    4 GTX 970/4-Way Sli

    I always have it on X16, the performance hit is like nothing to be concerned about really, not with today's gpu's.
     
  6. CrunchyBiscuit

    CrunchyBiscuit Master Guru

    Messages:
    343
    Likes Received:
    126
    GPU:
    MSI GTX 980 Ti
    In most scenarios, you can safely put it at 16x without a noticeable impact on performance.

    However, there will still be a noticeable performance impact during scenarios where your videocard is being stressed heavily. This occurs on all video cards due to the nature of how the feature works.

    It's rather easy to test the performance cost of AF for yourself. Find a spot in a game that really stresses the GPU and where you can barely meet your minimum target frame rate, preferably with lots of foliage and particles right in your face. Now change the AF settings and see the results.

    I just fired up Kingdoms of Amalur and stood around in a waterfall splash area (fillrate capped) for a quick demonstration:

    [​IMG]
    No AF - ranges from 47 to 50 fps.

    [​IMG]
    16xAF - ranges from 39 to 42 fps.

    About 15%-20% difference in performance on my potato PC. I think that's a pretty big impact, especially since I'm already way below my target frame rate of 50fps@50Hz.

    The difference only becomes obvious when the GPU is at its max fillrate, being stressed heavily. This also occurs on my brother's 980Ti and on a friend's 1050 in GTA 5. Some spots in the forest with lots of grass, shadows, ambient occlusion and 16x anisotropic can stress those cards pretty badly, and when you use a car to perform a burnout (drift in position), creating lots of dust, smoke and dirt particles, and position your camera in such a way that all those effects are visible at the same time, preferably as close to the camera as possible, the frame rate will dip (depending on the rest of the settings and the resolution used, of course). Now do the same thing with anisotropic filtering disabled and BOOM, +10 fps.

    AF just adds so much texture detail though, in most games I want it at 16x, despite the potential performance loss.
     
    BlindBison likes this.
  7. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    4,387
    Likes Received:
    627
    GPU:
    4 GTX 970/4-Way Sli
    Great find, I never paid attn. to the perfomance hit on af.

    I'll do some testing on my own, thanks for the find.
     
    BlindBison likes this.
  8. BlindBison

    BlindBison Ancient Guru

    Messages:
    2,404
    Likes Received:
    1,128
    GPU:
    RTX 3070
    @CrunchyBiscuit If what your saying there is correct across even modern high end cards, then this seems to be uncommon knowledge -- going off of what you're saying, I would expect the best "bang for your buck" setting might be 4x or 8x rather than 16x, but truthfully I don't know enough about the feature to say -- most seem to just set it and forget it globally at 16x, but I may do some experimenting with 4x or 8x passes. Personally I usually find it tough to tell the difference between 8x and 16x.
     
    Last edited: Jul 9, 2019
  9. CrunchyBiscuit

    CrunchyBiscuit Master Guru

    Messages:
    343
    Likes Received:
    126
    GPU:
    MSI GTX 980 Ti
    Hey again!

    It's all up to you (personal preference). If you can't find much of a visual difference between 8x and 16x in a certain game, feel free to go for 8x. In some games 16x is a must (for me), but in other games 8x looks fine as well. It also kinda depends on the mipbias the game applies (negative values require less anisotropic, but might exhibit more shimmering and pixel crawling) and what matters to you more - image quality or performance.

    Anyway, just enjoy your graphics and try not to let others dictate what settings you should be using. Experiment and go with what works for you. Game on and have fun!

    EDIT: I found a pretty relevant thread about exactly this topic, might be worth a read:
    https://www.neogaf.com/threads/anis...nes-is-16x-af-still-considered-cheap.1359136/
    Comments and in-depth tests by Barrow Roll later in the thread are interesting.
     
    Last edited: Jul 9, 2019
    BlindBison likes this.
  10. alanm

    alanm Ancient Guru

    Messages:
    12,222
    Likes Received:
    4,408
    GPU:
    RTX 4080
    First thing I do after driver install is set AFx16 globally. The perf impact is negligible (maybe 1-2%). I recall getting into a debate with someone over this some 10 years ago and showed him before and after 3dmark scores with AF off vs x16. It was as stated, negligible. Try it yourself, run some game or graphics benchmarks and see results.
     

  11. BlindBison

    BlindBison Ancient Guru

    Messages:
    2,404
    Likes Received:
    1,128
    GPU:
    RTX 3070
    If I force Anisotropic Texture Filtering to 8x or 16 x in the control panel, does it replace or add to the in-game setting?

    My understanding is that it should override the in-game setting, not add to it, but I've yet to find a definitive citation on that so I figured I'd double check here.

    @alanm Yes, what you're saying there appears to be the general consensus -- from what I gather, provided one has a GPU with a massive amount of VRAM and you're not playing at 4K, many recommend simply maxing out the "Texture Quality" setting since its normally VRAM dependent + setting Anisotropic Filtering to 16x globally in the control panel and setting your polling rate to 1000 Hz since the perf impact is apparently negligible for that as well.

    However, inline with @CrunchyBiscuit 's findings above and that thread he linked, it does appear that in some cases, AF can have a meaningful performance impact (thought these cases may not be the norm). I think it may be notable that a very small performance impact of 1-2 fps may not be negligible if you can't tell the difference between 8x and 16x passes for example. Arguably the same principle could be applied to 500 hz polling rate vs 1000 hz, but the perf impact there may be even smaller than 1-2 fps -- not sure there.
     
  12. tsunami231

    tsunami231 Ancient Guru

    Messages:
    14,702
    Likes Received:
    1,843
    GPU:
    EVGA 1070Ti Black
    That was one worst examples of AF off vs on I seen, (The enlarged parts of the images look worse on then off) show perf hit sure, but if the gpu is already at full load with it off there will be hit putting it on i mean the card is already maxed out. thats just common sense.


    [​IMG]


    [​IMG]

    Skyrim has broke implementation of AF even when it on it looks off or broke, I usual Force AF in nvcp on game by game basis if said games dont use AF or it usage of it is broke. for most part I see zero perf hit with it on, I still game at 1080p 60fps if I try to run 4k or higher then there is already huge hit to perf that AF hit is irrelevant. Texture quality is set highest in cp though.

    If the gpu is maxed out already and af is off turning and you not caped at your framerate turn it on is gonna have hit. if you are max gpu usage and you are at framerate cap then you will probably never notice hit, cause there still some leeway there is no leeway if you have it off and you dont even have your framerate capped turning it on is just gona drop more. AF perf hit negligible compared to other things like AA or AO.


    Not sure what this poll rate talk is about? we talking about the mouse poll rate? that should have zero barring on performance in any way these days maybe in the days of single or dual core but not now, obviously yes there would hit if said system is already maxed out cpu/gpu wise.

    Again if something is maxed out be it GPU or CPU turning on more effects and or features are not gone make run better it just make run worse depending on the effect
     
    Last edited: Jul 9, 2019
    BlindBison likes this.
  13. RealNC

    RealNC Ancient Guru

    Messages:
    4,894
    Likes Received:
    3,168
    GPU:
    RTX 4070 Ti Super
    @StephenEnzor

    Personally, I haven't encountered a game where 16x NVCP forced makes a significant difference. It's always something like a 1-3FPS difference. But obviously I didn't play every game in existence :)

    It is possible that some games could see a significant performance difference. Not only because of 16x, but because it's forced by the driver on all textures. If a game uses some weird rendering tricks and for some reason uses an exceptional amount of textures to achieve some effect, those textures will be marked by the developer to not get any filtering at all. The driver however will apply AF unconditionally, and thus with huge amounts of textures this would see a perf impact.

    Visually, in most games the graphical improvement with in-game AF vs NVCP forced AF is very minor. But in some cases it's a rather huge improvement. Some racing games for example will still have blurry road lines even with the highest in-game AF setting. Forcing 16x or even just 8x in NVCP fixes the blur in most cases. In general, the improvement this can make varies a lot from game to game, from no difference at all (the game already has good AF settings for its textures) up to a pretty night and day difference. In some rare cases though, NVCP AF can result in glitches. Race Driver Grid (the first one) will have white shimmering lines with NVCP forced AF. I think I encountered another game with such issues, but I can't remember which one.

    Lastly, I've heard of some games having an in-game texture filter setting that is mislabeled. You set it to the highest value for example, but instead of this just doing 16x AF, it does more than that (like anti-aliasing or post-processing.) But this has nothing to do with NVCP forced AF though.
     
    Last edited: Jul 18, 2019
    BlindBison likes this.
  14. BlindBison

    BlindBison Ancient Guru

    Messages:
    2,404
    Likes Received:
    1,128
    GPU:
    RTX 3070
    @RealNC Thanks, that's helpful!
     
  15. Caesar

    Caesar Ancient Guru

    Messages:
    1,555
    Likes Received:
    680
    GPU:
    RTX 4070 Gaming X


    Performance-wise, AF isn’t too computationally expensive.

    ....a typical texture is far too computationally expensive to unconditionally reuse in a scene because the relative distance between the texel (a pixel of a texture) of the object and the camera affects the observable level of detail, which could easily translate to wasted processing time spent on obtaining multiple texture samples that are applied to a disproportionately small surface in the 3D scene. .....

    .... when a texel is trapezoidal there is insufficient sampling in both directions. To solve this, anisotropic filtering scales either the height or width of a mipmap by a ratio relative to the perspective distortion of the texture; the ratio is dependent on the maximum sampling value specified, followed by taking the appropriate samples. AF can function with anisotropy levels between 1 (no scaling) and 16, defining the maximum degree which a mipmap can be scaled by, but AF is commonly offered to the user in powers of two: 2x, 4x, 8x, and 16x. The difference between these settings is the maximum angle that AF will filter the texture by. For example: 4x will filter textures at angles twice as steep as 2x, but will still apply standard 2x filtering to textures within the 2x range to optimize performance. There are subjective diminishing returns with the use of higher AF settings because the angles at which they are applied become exponentially rarer....

    Source: https://www.geforce.com/whats-new/guides/aa-af-guide#1
     
    Last edited: Jul 18, 2019
    BlindBison likes this.

  16. BlindBison

    BlindBison Ancient Guru

    Messages:
    2,404
    Likes Received:
    1,128
    GPU:
    RTX 3070
    @Caesar Thank you very much -- going off of this line:

    4x will filter textures at angles twice as steep as 2x, but will still apply standard 2x filtering to textures within the 2x range to optimize performance.

    This may be why performance remains reasonable I expect -- because the driver is not applying 16x to every texture all the time, it only does it when it's needed if I'm understanding correctly while applying lower values when they would apply.

    I don't know, perhaps I'm misunderstanding, but since the perf impact seems to be very low going off what people are saying here, I'll probably just force it to 16x in the driver.
     
  17. Apparatus

    Apparatus Master Guru

    Messages:
    429
    Likes Received:
    73
    GPU:
    3080 ti Vision OC
    I believe that this way you can't be 100% sure that you have AF properly or even applied in some cases.

    Your strategy should work in 95% of games.
    But sometimes you can achieve better in terms of optical result and framerate with the built-in AF.

    It's also true that Nvcpl normally overrides the built in setting.
    But there are some rare cases where you have to disable the built-in setting in order to achieve proper implementation or to avoid conflicts.

    I can't give you examples, but I remember to deal with such scenarios back in my OCD and time abundant days of video gaming:)
    I used to make screen shot comparisons and framerate measurements after the installation of a game, and then deciding my AF(same for AA) strategy to follow.
     
    BlindBison likes this.
  18. BlindBison

    BlindBison Ancient Guru

    Messages:
    2,404
    Likes Received:
    1,128
    GPU:
    RTX 3070
    @Apparatus Thank you very much, that's helpful to know -- in that case, I expect it may be preferable to leave the setting on "Let the App decide" then manually set it for games that don't have a decent implementation or that don't let you decide on the filtering amount in-game.
     
  19. A M D BugBear

    A M D BugBear Ancient Guru

    Messages:
    4,387
    Likes Received:
    627
    GPU:
    4 GTX 970/4-Way Sli
    Normally I have it under nvidia inspector: Application settings, then set the game to x16, at times I might force it under inspector, but I am alot more highly interested in AA image quality such as sparsegrid and hybrid aa, that's where its at, overall image quality, I always big into this for a very long time.

    But far as the performance hit on af, seems rather interesting, have to conduct this on my own side, thanks alot.
     
  20. Loobyluggs

    Loobyluggs Ancient Guru

    Messages:
    5,210
    Likes Received:
    1,583
    GPU:
    RTX 3060 12GB
    Digital Foundry have biased reporting, and I simply will never trust anything they 'report' on.

    OT: I always max AF, always have, it really is a left-over from the 90's that doesn't appear to do anything if you set it low or high, so, high - if I ever get a game or application that contradicts me on this, I'll let you know.
     

Share This Page