Methodic testing of gfx settings versus fps / gpu load: some observations

Discussion in 'Technical Archives' started by Niels_at_home, Jan 19, 2012.

  1. Niels_at_home

    Niels_at_home Registered

    Joined:
    Dec 30, 2010
    Messages:
    21
    Likes Received:
    1
    Hello!

    I have been testing various graphics settings and making a note of the GPU usage and framerate and noticed a few things.

    My PC:
    Intel Core 2 duo 3.8Ghz
    Ati 5870 2GB default clocks
    Windows 7 32bit

    Test Situation:
    1920x1080 zero AA
    F3 at Spa practise no AI
    Cockpit view, no HUD, no arms/wheel, default VOF
    Frame rate taken just sitting in the pits stationary
    The CPU use by rFactor was never maxed out. rFactor topped at some 65%, total CPU use never much higher.

    Observation 1: Mirrors affecting framerate and 'second time pit out' difference LOW DETAILS

    With these detail settings:

    Circuit Low
    Player Low
    Opponent Low
    Texture Low
    Filter x4
    Effects Off
    Shadows Med
    Blur Off
    Movies Off
    Occlusion No
    Drops No
    R Refl No
    E Refl No
    Motion No

    Mirrors OFF
    First time out pits, 260 fps, GPU 83%
    Second time out pits, 260 fps, GPU 83%

    Mirrors ON
    First time out pits, 150 fps, GPU 55%
    Second time out pits, 190fps, GPU 70%

    I get this all the time; first time enabling mirrors is a huge fps hit, but escape, and back out on the track, I gain 40fps.

    However!
    Setting shadows OFF (instead of MEDIUM):

    Mirrors OFF
    First time out pits, 245 fps, GPU 63%
    Second time out pits, 245 fps, GPU 63%

    Mirrors ON
    First time out pits, 205 fps, GPU 63%
    Second time out pits, 205 fps, GPU 63%

    So with shadows at Medium, there is no difference going out and in the car again. Note that NO shadows NO mirrors gives a lower FPS than MEDIUM shadows and NO mirrors! :S


    Observation2: Higher details

    Circuit High
    Player Low
    Opponent Low
    Texture High
    Filter x8
    Effects Off
    Shadows Med
    Blur Off
    Movies Off
    Occlusion No
    Drops No
    R Refl No
    E Refl No
    Motion No

    Mirrors OFF
    First time out pits, 215 fps, GPU 81%
    Second time out pits, 215 fps, GPU 81%

    Mirrors ON
    First time out pits, 134 fps, GPU 57%
    Second time out pits, 168 fps, GPU 73%

    Still a strange difference doing the 'in car, escape, in car again' routine when using mirrors.


    Observation 3: Higher details with high player detail

    Circuit High
    Player high
    Opponent Low
    Texture High
    Filter x8
    Effects Off
    Shadows Med
    Blur Off
    Movies Off
    Occlusion No
    Drops No
    R Refl No
    E Refl No
    Motion No

    Mirrors OFF
    First time out pits, 171 fps, GPU 65%
    Second time out pits, 162 fps, GPU 64%

    Mirrors ON
    First time out pits, 132 fps, GPU 58%
    Second time out pits, 148 fps, GPU 66%

    Second time out pits, without mirrors, gave a slightly LOWER fps (tried it many times). Mirrors on, as in earlier tests, showed improved FPS on the second time out into the cockpit.

    Observation 4: Much higher details

    Circuit full
    Player full
    Opponent full
    Texture full
    Filter x8
    Effects full
    Blur on
    Movies Off
    Occlusion yes
    Drops No
    R Refl yes
    E Refl yes
    Motion No

    Shadows MAX:
    Mirrors OFF
    First time out pits, 67 fps, GPU 61%
    Second time out pits, 67 fps, GPU 61%

    Mirrors ON
    First time out pits, 57 fps, GPU 55%
    Second time out pits, 59 fps, GPU 56%

    Shadows MEDIUM:
    Mirrors OFF
    First time out pits, 71 fps, GPU 58%
    Second time out pits, 70 fps, GPU 58%

    Mirrors ON
    First time out pits, 60 fps, GPU 53%
    Second time out pits, 62 fps, GPU 54%

    The 'second time out' difference is much smaller, as if other unoptimized things are the main culprit here.


    Thoughts..
    Only the absolute lowest detail managed to get some 83% GPU load, as details increased, GPU load decreased and frame rates dropped perhaps more than you could 'explain' by the looks of the graphics. As if things hold it back, make it 'wait', which gets worse as more detail is added, which then of course progressively worsens the framerate. Extra detail is more work to draw, and seemingly done less efficiently.

    The mirrors framerate increasing 2nd time out of the pits is very weird, sounds almost like the rFactor 1 issue where system ram (instead of gfx ram) was used for things, slowing things down a lot.

    Perhaps these observations can point to certain things for the devs, perhaps not, in which case I wasted a few hours! :)
     
  2. F2kSel

    F2kSel Registered

    Joined:
    May 28, 2011
    Messages:
    138
    Likes Received:
    2
    Where are you turning the mirrors off, I've tried doing it at the race track and haven't noticed any great FPS difference, maybe 2-3 fps at best and usually only when cars are behind.
     
  3. Atle Dreier

    Atle Dreier Registered

    Joined:
    Jan 14, 2012
    Messages:
    84
    Likes Received:
    2
    Very interesting that GPU utilization goes DOWN as detail increase. I see the same with CPU, where adding AI DECREASE the CPU load. Something is bottlenecking BADLY in this software, no doupt about it. I can't wait to try it once they figure it out, it will be GORGEOUS and excellent!
     
  4. DocJones

    DocJones Registered

    Joined:
    Jan 26, 2011
    Messages:
    403
    Likes Received:
    4
    Interesting test and results. Hadn't heard about the 2nd time out higher fps issue. Will definately try that aswell.
    I did some tests regarding grafics card load myself and found that the more details ingame activated, the higher gfx load. Numbers constantly increased from <60% load with very low settings to 85% max load with very high settings (higher settings than playable framerate though). Very weird to see it the other way around.

    I hope ISI can make some sence out of it and can deliver a fix for the low grafics card load issue.
     
  5. F2kSel

    F2kSel Registered

    Joined:
    May 28, 2011
    Messages:
    138
    Likes Received:
    2
    I actually had the opposite at one time, second time out lower FPS although I haven't seen that for a while.

    I still get 2xFPS when I alt-tab to desktop and back but game plays worse which also makes it hard to monitor.

    Also check the second lap as that's were my FPS drop by half in some corners with less cars around.
     
  6. Hybrid

    Hybrid Registered

    Joined:
    Dec 27, 2011
    Messages:
    58
    Likes Received:
    0
    As always a intresting post from Niels.
    Niels cant you help out ISI abit with the FFB to get it even more top notch? ;)
     
  7. Atle Dreier

    Atle Dreier Registered

    Joined:
    Jan 14, 2012
    Messages:
    84
    Likes Received:
    2
    I find that the FFB is very nice once you get decent framerates. Mine is 'ok', but triplescreen is unuseable atm.
    I will conduct a similar test to Niels' and report back.

    I propose we make a common replay available and record min,med and max fps in different configs. That way we have a common ground to base performance on.
     
  8. Niels_at_home

    Niels_at_home Registered

    Joined:
    Dec 30, 2010
    Messages:
    21
    Likes Received:
    1
    Off topic, I think rF2 has a lot of potential; it does need a huge amount of 3d optimizations which I only hope they manage.. And no info is available on the tire model yet (apart that I'm sure its very clever, hopefully not TOO clever..).. At some stage I hope to get into the physics and FFB but that might be a longish time yet. :)
     
  9. Duvel

    Duvel Registered

    Joined:
    Dec 31, 2011
    Messages:
    332
    Likes Received:
    22
    As someone said, how are you turning off the mirrors? just the button while in the pits?
     
  10. Niels_at_home

    Niels_at_home Registered

    Joined:
    Dec 30, 2010
    Messages:
    21
    Likes Received:
    1
    Yup button while in pits.
     
  11. F2kSel

    F2kSel Registered

    Joined:
    May 28, 2011
    Messages:
    138
    Likes Received:
    2
    My tests with mirrors all seem to behave as expected. The less work they have to do the more FPS I get.
    I get less of a difference when AI are on the track, down to just two or three FPS difference.

    Shadows had no unexpected effects on frame rate.
     

Share This Page