Thinking about improving FPS, a path to optimizing?

Discussion in 'General Discussion' started by Emery, Oct 19, 2017.

  1. WiZPER

    WiZPER Registered

    Joined:
    Oct 5, 2010
    Messages:
    1,517
    Likes Received:
    186
    Sorry, but shaders don't care about the status of the visibility - not going to comment on the other part, we'll just have to wait and see :)
     
  2. demerzel

    demerzel Registered

    Joined:
    Aug 30, 2015
    Messages:
    197
    Likes Received:
    127
    No need to comment on. According to your statement, all shaders applying all the time because of reasons and due to this the game will render all sort of rain-related things in dry condition for no reason other than shaders don't care about the status of the visibility.

    Now the last thing I don't understand is: How come the McLaren performance fps wise was equally bad without the damage improvements?

    Probably shaders don't care about the status of their implementation too.
     
  3. WiZPER

    WiZPER Registered

    Joined:
    Oct 5, 2010
    Messages:
    1,517
    Likes Received:
    186
    Are you sure it's the last thing you don't understand ? :)
     
  4. demerzel

    demerzel Registered

    Joined:
    Aug 30, 2015
    Messages:
    197
    Likes Received:
    127
     
  5. WiZPER

    WiZPER Registered

    Joined:
    Oct 5, 2010
    Messages:
    1,517
    Likes Received:
    186
  6. Travis

    Travis Registered

    Joined:
    Dec 7, 2012
    Messages:
    245
    Likes Received:
    255
    Latest GT3 update benchmarks based on the following PC specs:

    Win 10 Pro x64 version 1709 (AKA Creators Update)
    OC'd i5 4670K @4.3GHz (Asus MB controlled overclock)
    Asus ROG GTX 1070 stock w/Nvidia 388 drivers. GPU utilising PCIe 3.0 @ 2.0 x 16
    16GB RAM @ 2400
    rF2 installed on a 500GB Samsung 850 EVO SSD

    Starting from 20th position, 19 AI cars ahead of my car. Default weather and TOD for Matsusaka.

    • Matsusaka 20 McLaren GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 106 FPS
    • Matsusaka 20 Bentley GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 98 FPS
    • Matsusaka 20 Merc GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 92 FPS
    • Matsusaka 20 Radical GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 85 FPS
    • Matsusaka 20 Calloway GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX medium: 85 FPS

    Using rF2's maximum graphical settings for science!:
    • Matsusaka 20 McLaren GT3 cars, start grid, max graphics settings, 1080p, AA level 2, post FX medium: 57 FPS
    • Matsusaka 20 Nissan GT1 cars, start grid, max graphics settings, 1080p, AA level 2, post FX medium: 92 FPS (61% increase in frame rate).
    Latest rF2 build using DX9:
    • Matsusaka 20 McLaren GT3 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX N/A: 114 FPS
    • Matsusaka 20 McLaren GT3 cars, start grid, max graphics settings, 1080p, AA level 2, post FX N/A: 66 FPS
     
    Last edited: Oct 25, 2017
  7. WiZPER

    WiZPER Registered

    Joined:
    Oct 5, 2010
    Messages:
    1,517
    Likes Received:
    186
    Are you starting from same grid position ?
     
  8. Travis

    Travis Registered

    Joined:
    Dec 7, 2012
    Messages:
    245
    Likes Received:
    255
    Yes.

    Every time I run a test, I'm using 19 AI cars or 20 including my own, from the back of the starting grid. I'm using the default weather & time of day (TOD).

    I'd be interested reading other users findings to be honest.
     
    WiZPER likes this.
  9. stonec

    stonec Registered

    Joined:
    Jun 19, 2012
    Messages:
    3,399
    Likes Received:
    1,489
    Well, my findings changed even less than yours on the new version:

    Radical GT3 v100: Matsusaka, 20 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX low: 54 FPS
    Radical GT3 v109: Matsusaka, 20 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX low: 55 FPS

    Meanwhile:

    McLaren GT3 v179: Matsusaka, 20 cars, start grid, fresh profile graphics settings, 1080p, AA level 2, post FX low: 65 FPS

    So there is 1 FPS change with updated LODs for me, which might as well be an artifact. Meanwhile, difference between McLaren and Radical is 18% in FPS. So whatever makes some cars heavier on FPS than others, it's not due to poor LODs. Looking at the car model, you would think that McLaren would be heavier than Radical, but it's the opposite, and with a big margin.
     
  10. Andregee

    Andregee Registered

    Joined:
    Jun 23, 2012
    Messages:
    927
    Likes Received:
    390
    That's still a big problem I think. Undepending of the API the engine seems to be at the limit and can't handle higher polygone modells in a proper way. Maybe it was a dead end to use the old engine to keep the compability with old content. Maybe S397 is working on an engine overhaul that will fix the problem so that's why they are keeping the path of high load content
     
    joekart likes this.
  11. joekart

    joekart Registered

    Joined:
    Dec 24, 2011
    Messages:
    55
    Likes Received:
    33
    Yes Andregee, i think and hope the same.
     
  12. bobbie424242

    bobbie424242 Registered

    Joined:
    Oct 17, 2017
    Messages:
    273
    Likes Received:
    357
    Here's my results after some benchmarking.

    Win 10 Home x64 version 1709 (AKA Fall Creators Update)
    Core i7 860 / 16GB RAM / NVIDIA GTX 970 overclocked, drivers 388.00
    rF2 installed on a 500GB SSD

    Graphic settings:

    D3D9, 1920x1200
    in-game AA settings overridden by NVIDIA Inspector with 8xQ/2xSGSSAA
    vsync off (for unlocked framerates)

    In-game graphic settings:

    Circuit detail: full
    Player detail: full
    Opponent detail: medium
    Texture detail: full
    Texture filter: 16x aniso
    Special effects: full
    Shadows: high
    Shadow blur: fast
    Soft particles: high
    Rain drops: yes
    Road reflection: low
    Environment reflection: off
    Visible vehicles: 12


    Test settings:

    Cockpit: seat position 0/0, wheel and mirrors disabled
    Zandvoort default weather and TOD
    8 AI cars identical to tested car
    Tested car in pos #9, so with 8 AI in front

    All fps numbers are at the race standing start grid and fps is cockpit / hood / bumper. All conditions are the same except the car changing:

    Radical GT3: 46 / 50 / 54
    McLaren GT3: 56 / 60 / 67
    Mercedes GT3: 53 / 53 / 60
    Honda Civic BTCC: 73 / 75 / 82
    Megane Trophy: 53 / 53 / 60

    So we can see the GT3 still have lower fps and with varying numbers between GT3's, the Radical
    having clearly lower fps.
    The Civic has way higher fps while the Megane (not a gt3) has the same numbers that the Merc GT3.
    So performance is highly car dependent with a lot of fluctuation, which is not exactly great
     
    Last edited: Oct 25, 2017
    M D Gourley likes this.
  13. Ricknau

    Ricknau Registered

    Joined:
    Nov 12, 2011
    Messages:
    778
    Likes Received:
    39
    I don't see the "baked" shadows under the car when Shadows is set to low. I wish i did. It looks terribly sterile without shadows under the car. When I put shadows on medium my frame rate drops by half. This is with the 917 Porsche which is a huge frame killer anyway.

    Wonder what I need to do to get the baked shadows. I'm running the DX-11 beta.
     
    Guimengo likes this.
  14. WiZPER

    WiZPER Registered

    Joined:
    Oct 5, 2010
    Messages:
    1,517
    Likes Received:
    186
    The 917 Porsche may not be modded correctly then, can't really blame the title for that. If shadows are being rendered based on a high poly object, and not a shadow dummy, this will be the result.
     
  15. stonec

    stonec Registered

    Joined:
    Jun 19, 2012
    Messages:
    3,399
    Likes Received:
    1,489
    The reason that low shadows perform better is precisely because there is no dynamic shadow, there is only a simple static undercar shadow box applied, and problem is all mods don't come with that shadow box texture.
     
    WiZPER likes this.
  16. Filip

    Filip Registered

    Joined:
    Feb 9, 2017
    Messages:
    1,208
    Likes Received:
    937
    I put shadows on High because Low is unacceptable immersion killer for me, watching floaty cars is distracting. Medium looked bad too
    But Shadow blur turned off gives some FPS boost without loosing too much visually imo.
    Circuit details to Low is also good try
     
  17. Ricknau

    Ricknau Registered

    Joined:
    Nov 12, 2011
    Messages:
    778
    Likes Received:
    39
    Yeah I see that now. Tried a few more cars and there it is. Not real immersive but boy what a frame rate benefit. I wish I could use it with the 917 because that car itself really taxes my system. (Trying to maintain 90fps for VR) On most other cars I can get away with medium and stay at 90 even at the start of the race, but not the 917.
     
  18. dadaboomda

    dadaboomda Registered

    Joined:
    May 15, 2016
    Messages:
    712
    Likes Received:
    202
    Just to complete :

    1) Shadows is absolutely THE setting which is the worst in terms of performance in rf2. There is a huge lack of optimization. Really it is the worst. Just to say : other worst settings in terms of performance are : Environment reflection, PP, AA, real mirrors, number of AI. Very important concerning environment reflection and shadows : more there are trees (and objects reflecting and shadowing) more it is worst : it seems logical but we don't think about this. So for example, if you are like me a fanboy of the Nordschleife thousands of trees.... turn shadows and reflection to off. I will explain you why you can do this and having beautiful graphics on a small gtx 780 with perfect smooth.

    2) level of shadows and shadow blur : graphics are horrible and I am not a gamer for who graphics are really important. Why I say this : absolutely nothing combination of these 2 settings make beautiful shadows.... the only combination which is better is ultra + ultra BUT even this setting is relatively ugly : a lot of aliasing and some weird problems of LOD according to the distance. There is a solution for this aliasing which is the baddest thing in terms of performance for a GPU : the name is ~sparse grid supersampling. If you dont have a gtx 1080 ti, just forgot (perhaps I exagerate a little bit but...)

    Conclusion :
    - shadows = the worst performance of rf2
    - even with ultra + ultra settings, that's ugly. If you have a very high end GPU you can try to force sparse grid supersampling.

    Very simple question:
    is it worth to completely destroy the ability of the GPU to make the game smooth (a simulation must be completely smooth we agree ???).

    Answer:
    surely not. For me the shadows were indispensable for immersion on the nordschleife (thousands of shadows). But this is not really necessary for immersion (I was wrong and I took 5 years to understand it, Bravo David;)).

    Disabling the shadows (and in my case the reflections of the environment that I strongly advise to disable too because we do not see the difference when we drive or almost really not, and the post process for my little GPU that anyway are not successful - sorry S397 -)

    ---->>>>>> PERMITS TO ENHANCE IMPROVEMENTS IN MUCH MORE IMPORTANT DOMAINS:

    - change from 1K resolution to 4K resolution (via the nvidia DSR function by lowering the DSR blur setting (important).
    - Parameters of the GPU driver set to maximum quality

    CONSEQUENCE :
    You may not believe me but changing these few parameters ENORMOUSLY improve the overall graphic quality with a total smooth simply by deactivating the 3-4 most disastrous graphical parameters in terms of performance :

    ---- >>> resolution 4 times higher, better performance and especially FPS STABLE, a lot of detail in the whole scene, no flicker, a high quality image, aliasing extremely weak (level 1 for gtx 780) etc.

    Other parameters that must be disabled next: real mirrors, activate the "minimum framerate" which is not magic but works wonders if your vsync is 60 and the GPU can only display 58-59 ... And put less AI and low (yes you may have less elements on some AI but you prefer to have a full shift loss of FPS ???????

    Principles settings are ULTRA : track detqils, road detqils, anisotropic x16 etc.

    IMPORTANT :
    If your GPU is better than mine, dont desactivate all worst settings (PP or real mirrors for example). ALWAYS begin by desactivte shadows.
     
    Last edited: Oct 31, 2017
  19. stonec

    stonec Registered

    Joined:
    Jun 19, 2012
    Messages:
    3,399
    Likes Received:
    1,489
    You are free to deactive shadows, however, I disagree it has no visual consequence. TV cam shots in partiucular look ridiculous without shadows and cars are floating on the ground etc. 4K upsampling, I don't see the benefit for it myself. I own a 24 inch 1080p monitor. Do I really need 4 times the resolution? 4K will also require almost 4 times more GPU power than 2K, so it's not a small sacrifice. If I owned a GTX 1080, then sure, it would make sense to enable it.
     
  20. WiZPER

    WiZPER Registered

    Joined:
    Oct 5, 2010
    Messages:
    1,517
    Likes Received:
    186
    It prob don't have any LODs. Or is that just as private practice, no AIs ?
     

Share This Page