Are we Benchmarking yet?

Discussion in 'General Discussion' started by davehenrie, Apr 27, 2017.

  1. davehenrie

    davehenrie Registered

    Joined:
    Jul 6, 2016
    Messages:
    7,454
    Likes Received:
    4,369
    In anticipation of the DX11 update, how many of you are recording your current fps? And I don't mean casually observing the fps while driving. I mean actually logging it. Noting on paper or on the computer not only the fps but the cars, track time of day... You shoul get as detailed as possible... and also take multiple readings. Note the fps with the # of cars and which track at the start, at the end of lap 1 and then perhaps several more times.
    I suspect we will see many many posts that claim "MY SYSTEM ALWAYS GOT 150 fps ON EVERY TRACK and now with DX11, it is much worse/better.
    So write it down, as many tracks as you can, as many times as you can and as many different cars.
    ONe of you spreadsheet gurus could perhaps make a template????

    dh
     
  2. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    I've been doing a little bit of benchmarking recently, though not in direct response to impending DX11 change.

    Basically, one of the rF2 oddities I've noticed relates to AA level. With 0 AI, AA level has an obvious impact on fps. Add 15 AI and fps is virtually the same regardless of AA level. AA level should only be affected by GPU power, so this result is puzzling to me. This is on my triple 1080p screens powered by a GTX 1080.

    15 AI runs about 60% fps of 0 AI. 15 online opponents runs about 80% of 0 AI. Obviously this comparison should be dependent on CPU and my CPU is a 3.4 GHz i7.

    ***
    The fps hit for adding AI is where AC shines (less physics are being calculated or is it really that much better optimized?), as adding 15 AI runs about 90% of 0 AI. AA level has a similar fps effect regardless of number of AI.
     
    Last edited: Apr 27, 2017
  3. suryaprihadi

    suryaprihadi Registered

    Joined:
    Apr 16, 2017
    Messages:
    87
    Likes Received:
    32
    I spent sometime logging current DX9, hoping to understand more of why RF2 is not pushing my GPUs to higher than 70% in cockpit view. I do hope DX11 will improve things.

    While testing I discovered something which will be very useful, at least for me if not for the rest of the gang here.

    Achieving higher FPS is important and I see that minimum 60 FPS is what I need to get, 30 to 45 FPS in fast corners does look and feel bad. Since my chassis is still on its way, I cant install steering wheel, I have to use AI driving. Actually I can see better how the overall quality of the game with the AI driving for me and indeed one's driving experience can negatively be effected by low FPS. This is not factoring yet that per 1 FPS slower, your car positional accuracy will be impacted by "X" meters as others mentioned.

    However, my test educated me not about fast 144* FPS ( *in reference to 144hz monitor ) or higher per se , but the test showed that I also need to seek FPS consistency, more so if the monitors are triple 2560 x 1440 or higher resolution as my GPUs are not up to the task at 100% best setting.

    I am new to gaming on PC and RF2 is my first venture actually into PC gaming, but I know a bit about video editing and cameras. One thing I learnt about gaming is that, it is so GPU dependent, multi-cores CPU beyond 4 cores doesn't contribute . Since we drive the sim in real time, delay is not acceptable. Same like video camera, it takes more band-with to do fast panning in camera , it also take more GPU power when a car in RF2 enter and exit S corners, because the scenery changes so much, then the FPS get hammered down, if one's setting image quality is at 100% MAX. This is why, I now am looking at how to achieve FPS consistency.

    Too bad I am an IT dummy, if I am better with IT, maybe I can explain better technically, but here goes........

    Below is the data logging from Afterburner with 2 data points per second ( 500 ms ).
    NOLA C layout. GTR500 2013. AI driving from pit and complete 1 lap. So approx 2+ minutes.
    I choose NOLA and GTR500 2013 because it is one of the newest from RF2.
    Since it is AI driving, anyone wishing to bench their GPU with current DX9 can use the same parameters I use and see how better DX11 will be for your GPU and my GPU.

    Sorry, no chassis means I cant use my wheel....:(, so AI driving it is then.
    During the test the game list mentioned 10 AI, but at the most during the test, only 2 AI cars will be physically present and only 1 AI car will be very close to my own AI car. Just in case as Emery pointed out, total 'X" numbers of AI cars may matter to FPS.

    4K v1.JPG

    The massive FPS drop is the S corners #5 to # 9 and then #11 to #13, something like that :)

    [​IMG]



    In summary at 100% best resolution, Single 4K, Tripple 2560 and single 2560 monitors
    MIX - v1.JPG

    So based on my current GPUs, no way I can use all 100% best setting for Single 4K or Tripple 2560 monitors. The dip to as low as 30 FPS for single 4K and 27 FPS for triple 2560 is so visually obvious in the game and look like shi-et and I am sure I would no want to take corners at 30 FPS.


    Since my 4K monitor is only 60 Hz, the only way I can do minimum 60 FPS is to reduce image quality and I tried many things, and ended up choosing these reductions below for 4K. In the AI test there is no rain yet, so I do not know how much FPS reduction is actual rain occurs in the game.

    4K v2.JPG
    The initial slower FPS is from pit to track. In the track its all those faster FPS.
    Above image quality reduction produced FPS consistency.
    What I can not yet phantom is , and due to my poor IT 101, why my GPU can't be throttled up to higher load to compensate for those massive +-18 FPS drop ? Automobilista & Assetto Corsa can load my GPU to 99% easy, I then can lock my FPS to make GPU load less. I feel its a waste of my GPU processing power if not at least 85% loaded in RF2.

    Since I am new to this PC gaming, I do not yet have the skill and knowledge to pixel peep or visually identify WHERE those image quality reduction went to:D , it look nearly the same to me. Fortunately at the least I can tell visually, where FPS matters most in the game....for me, in relation to the driving quality.

    If for video editing, I can tell low bit rate render result vs high bit renders immediately. I know where to look and I know what does not look natural due to pixelization or poor color gradation of say a sun-ball and so on. PC gaming is all fake images from man made pixels, it look very low grade CGI as it is , so I can't tell where the reduction went too.:oops:


    If for my triple 2560 monitors and since it can do 144hz, I will seek 85-100 fps constant with current GPU.
    Too bad there is no simple on screen menu to lock FPS in RF2. I need to dig to its program file..correct ?

    I tried a few reductions....
    triple 2560 v1.JPG

    But I have to settle for this......
    triple 2560 v2.JPG


    Now single screen 2560 x 1440

    2560 v1.JPG


    2560 v2.JPG
    I sure hope DX11 will allow higher GPU load for my GTX980 in 2 way SLI and soon GTX1080Ti in 2 way SLI.

    Hope the data may be useful to some...
     
  4. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    In ..\UserData\player\player.JSON (edit with Notepad or similar) are the following lines:
    "Max Framerate":0,
    "Max Framerate#":"0 to disable (note: positive numbers only, we always use the 'alternate' method now)",

    Change 0 to the fps you wish to use as your maximum in the first line. The second line is a comment with no effect, because it is in quotes after the colon :)). It's in all the ISI-based games, some may provide a UI method of setting it and some don't.
     
    Last edited: Apr 30, 2017
  5. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    And I sure hope you post tests of the GTX1080 Ti in SLI at these resolutions once they arrive!
     
  6. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    @suriyapiradi -
    I did a quick fps check with Fraps so you'd have a similar comparison. Will try to update with Afterburner charts if I get a chance.

    Single GTX 1080, mostly max settings, 37 degree FOV (also tested 60 degree FOV, but FOV's effect was insignificant), Practice session
    • 1920x1080 windowed
      • Avg 229 fps
      • Min 197 fps
      • Max 265 fps
    • 5760x1080, multiview disabled (e.g. one wide flat screen)
      • Avg 142 fps
      • Min 126 fps
      • Max 158 fps
    • 5760x1080, multivew enabled (e.g. side monitors angled in)
      • Avg 116 fps
      • Min 105 fps
      • Max 127 fps
     
    Last edited: May 1, 2017
    suryaprihadi likes this.
  7. suryaprihadi

    suryaprihadi Registered

    Joined:
    Apr 16, 2017
    Messages:
    87
    Likes Received:
    32
    Wow, there is an effect between multiview enabled and disabled ?

    I use default FOV, I dont know how many degrees that is, but it is narrow. Wide FOV actually has a bit of effect , a bit slower FPS if at say at 85 degrees, but not much though.

    If you can replicate my test parameters on your GTX1080 with 1 x 1080 monitor windowed at 100% best max and at my 2nd test parameters of 2650 x 1440 and get Afterburner to log , I will do the same with my 1080p monitor and we can see single GTX1080 vs GTX980 in 2 way SLI. I am sure your GTX1080 will do better being more powerful and not in SLI. 2 way SLI is never 200% equivalent most of the time.
    I only have now 1 of 1080p monitor, so I can't test triple 1080p multiview.

    At the rate my SLI GPUs being driven only 65% load x 2 at best, I am sure your GTX1080 will be better , as I think it is already 50+% more powerful than a single GTX980, based on this test below :
    http://gpu.userbenchmark.com/Compare/Nvidia-GTX-980-vs-Nvidia-GTX-1080/2576vs3603

    It will be fun and useful to see the result.:D
    This RF2 test thingy is also my first venture into SLI. My Vegas Pro video editor wont recognize SLI and it can't use newer GPU than the oldie 2012 GTX580 Fermi architecture for superb GPU render assist.
    So your test and mine will be a good database for RF2 potential SLI users

    If there is a way Assetto Corsa can do AI driving, we can bench AC too as it is DX11 and it can push my SLI to 99% x 2 load.

    Thanks...
     
  8. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    Sadly, rF2 did not like Afterburner when I tried it, just like AC doesn't like Fraps.
     
    David Kolody likes this.
  9. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    Yes, because the game is rendering three separate views rather than one. To me, it's amazing it is 50-60% of the single screen rather than only 33%.

    AC has a built-in benchmarking tool. Just click the button in the video setup, same location where you choose triple or single screen.
     
    Jake Loewenstein likes this.
  10. suryaprihadi

    suryaprihadi Registered

    Joined:
    Apr 16, 2017
    Messages:
    87
    Likes Received:
    32
    Okey dokey...

    Yes, I just realized AC has benchmarking tool...thanks.

    I think multiview wing monitors are the only ones angled, so middle one lower load ...so not 33% but can get 50-60% .. :)
     
  11. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    For Silverstone GT Layout and GT500, Practice session
    Single GTX 1080, mostly max settings, 37 degree FOV
    • 1920x1080 windowed
      • Avg 221 fps
      • Min 188 fps
      • Max 253 fps
    • 5760x1080, multiview disabled (e.g. one wide flat screen)
      • Avg 144 fps
      • Min 122 fps
      • Max 169 fps
    • 5760x1080, multivew enabled (e.g. side monitors angled in)
      • Avg 114 fps
      • Min 94 fps
      • Max 136 fps
     
    Last edited: May 1, 2017
  12. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    For VLM Sebring 12h and GT500, Practice session
    [worth noting that Race session is 15-17% lower than as tested in Practice session due to extra objects for Race]
    Single GTX 1080, mostly max settings, 37 degree FOV
    • 1920x1080 windowed
      • Avg 133 fps
      • Min 106 fps
      • Max 190 fps
    • 5760x1080, multiview disabled (e.g. one wide flat screen)
      • Avg 130 fps
      • Min 105 fps
      • Max 186 fps
    • 5760x1080, multivew enabled (e.g. side monitors angled in)
      • Avg 86 fps
      • Min 66 fps
      • Max 115 fps
     
  13. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    For Matsusaka International and GT500, Practice session
    Single GTX 1080, mostly max settings, 37 degree FOV
    • 1920x1080 windowed
      • Avg 196 fps
      • Min 172 fps
      • Max 218 fps
    • 5760x1080, multiview disabled (e.g. one wide flat screen)
      • Avg 124 fps
      • Min 106 fps
      • Max 149 fps
    • 5760x1080, multivew enabled (e.g. side monitors angled in)
      • Avg 98 fps
      • Min 84 fps
      • Max 113 fps
     
  14. muz_j

    muz_j Registered

    Joined:
    Aug 20, 2014
    Messages:
    264
    Likes Received:
    188
    if you're running it with rivatuner's statistics server for the OSD, then update RTSS to the latest beta version - v7. That resolved my issue with rF2 failing to start on the latest Win10 (Creator's Update).
     
    Emery likes this.
  15. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    Thanks, I'll try that.
     
  16. suryaprihadi

    suryaprihadi Registered

    Joined:
    Apr 16, 2017
    Messages:
    87
    Likes Received:
    32
    I only test in multiview/multiscreen mode, that angled wing monitors.

    SilverStone.JPG
    Looks like the SilverStone GT track is a bit lighter than Nola-C , or RF2 allows my GPU to do more work.
    My GPU2 in Silverstone GT is allowed to do 71%ish load , while in Nola-C only 65% ish. GPU 1 is also allowed to work 5% harder in Silverstone GT.


    DX11 Assetto Corsa benchmark feature which allowed my GPU to 99% and all ULTRA/BEST, is not too bad FPS wise compared to RF2 , but I don't want to run 99% GPU, there is no headroom for other task.
    Only in AC I am seeing my GPU memory max out at 4.07GB...holy cow !! In RF2 at best I seen 3.4 - 3.7 GB or so. 50FPS is still not nice to watch. Not so fluid I would say.

    Sorry 7680 x 1440 ...LOL
    Assetto C Bench.JPG

    What I cant understand is, why I don't see 25FPS-ish dip as bad as RF2 for AC in ULTRA/MAX ?
    The AC track has many corners too or fast scenery change and more AI cars visible too.

    End of Test
     
  17. WiZPER

    WiZPER Registered

    Joined:
    Oct 5, 2010
    Messages:
    1,517
    Likes Received:
    186
    Fast scenery change is not the general track render problem, long straights are since they require higher LOD values. Sebring as an example is particulary tough as it has two straights on each side of the pits and paddock area AND several places where two parts of the track are on either side of a wall.
     
    suryaprihadi likes this.
  18. suryaprihadi

    suryaprihadi Registered

    Joined:
    Apr 16, 2017
    Messages:
    87
    Likes Received:
    32
    Yes yes, that I notice at NOLA-C near the finish line where it has a long wall, like 5FPS lost to that. So at Nola-C on the corners I assumed, must be the changing surrounding features/buildings yah...not only the track itself. Good to know, thanks Wiz.
     
  19. Rui Santos

    Rui Santos Registered

    Joined:
    Jan 8, 2012
    Messages:
    1,081
    Likes Received:
    1,206
    Test at night, Suzuka for example, with artificial lights, FPS drop like hell!!
     
  20. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    3,035
    Likes Received:
    1,654
    I'm done fiddling with the DX11 beta... just can't get it up to the desired fps. AA level has no effect, post processing level has no effect. Cleared cbash & shaders. Thought I'd found fps by renaming player.JSON, but I was mistaken.

    NOLA-A, GT500
    Single GTX 1080, mostly max settings, 37 degree FOV, Practice session, Level 5 AA & Ultra
    • 1920x1080 windowed
      • Avg 96 fps, was 229 fps
      • Min 81 fps, was 197 fps
      • Max 108 fps, was 265 fps
    • 5760x1080, multivew enabled (e.g. side monitors angled in)
      • Avg 55, was 116 fps
      • Min 50, was 105 fps
      • Max 61, was 127 fps
    So, I'll be reverting to DX9 very soon. The visuals are very good in DX11, though corner workers are blooming in Ultra, so I'd probably choose High to avoid that.
     

Share This Page