[Official] DX9 vs DX11 performance difference

Discussion in 'Technical & Support' started by 4thworld, May 4, 2017.

Thread Status:
Not open for further replies.
  1. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    2,785
    Likes Received:
    1,378
    Looking at a lighter fps load rather than my stress test by using Palm Beach.

    settings: Level 4 AA, PPE=off, resolution 5760x1080 multiview, no sync, aggressive threads=off
    system: single GTX 1080, 8 GB VRAM; i7-4770 @3.4 GHz, 32 GB RAM
    circuit: Palm Beach
    car: AC Cobra
    view: cockpit camera
    weather: overcast

    DX9, Avg 109 fps, Min 100 fps, Max 122 fps
    Build 1813052, Avg 110 fps (+0%), Min 102 fps (+2%), Max 125 fps (+2%)
    Build 1828309, Avg 112 fps (+2%), Min 103 fps (+3%), Max 127 fps (+4%)

    In other words, on a less stressful test, we're getting gradual improvements with the new builds.
     

    Attached Files:

  2. juanchioooo

    juanchioooo Registered

    Joined:
    May 16, 2016
    Messages:
    1,205
    Likes Received:
    1,165
    In the interior camera of the vehicle goes quite well all to full 70 80 fps, but since you go outside as a viewer the fps fall from 70.80 to 30 fps, graphic card gtx 1060
     
  3. SPASKIS

    SPASKIS Registered

    Joined:
    Sep 7, 2011
    Messages:
    3,155
    Likes Received:
    1,420
    I just finished a set of tests with B1098 at Monaco. As I suspected, whatever is hurting fps at trackside view is not dx11 related. It can be reproduced as well with B1098. At the end of the post I have attached the different values obtained in 12 benchmarks.

    Description of the problem
    If I swap to trackside cam (AvPag) immediately after pressing race, benchmark fps figures (avg, min and max) show best results. I did up to 3 consecutive benchmarks with 85 fps average.

    However, if I press race, drive the full lap with cockpit view before switching to trackside camera, fps values significantly drop.

    Relevant Additional Info
    • The % of drop is variable, being difficult to understand why in some cases is bigger than others. After maintaining cockpit view for a few seconds in the pitlane, the drop was almost zero. However, worst results were obtained after one third of a lap with cockpit view before swapping. Quite worse than after a full lap!
    • All the benchmarks are performed using fraps with AI driving and time accelerating the outlap until a few seconds prior to crossing the FL where the benchmark is started and finished one full lap later.
    • Normal fps is restored every outing. This helps a lot in order to reproduce the issue.
    It would be interesting if "others" could try to reproduce this behaviour in their systems. It could be an indicator of something working faultly.
    I wonder why do I get higher framerate using trackside cams compared to cockpit view. Maybe whatever is happening with cockpit view that ruins trackside performance maybe also be ruining cockpit performance itself.

    I will try to check if using other driving camera shows different performance. Hopefully as good as trackside fps when working correctly. Later I might try latest beta. IMO these type of things that could provoke confusion when comparing different builds should be studied and sorted urgently.
    upload_2017-5-17_22-23-46.png
     
    Last edited: May 17, 2017
    Emery likes this.
  4. SPASKIS

    SPASKIS Registered

    Joined:
    Sep 7, 2011
    Messages:
    3,155
    Likes Received:
    1,420
    Benchmarking driving views with B1098. Definitely something is hurting cockpit view.
    hood w/o mirrors
    2017-05-17 22:28:00 - rFactor2
    Frames: 9054 - Time: 113469ms - Avg: 79.793 - Min: 61 - Max: 126
    hood with mirrors
    2017-05-17 22:31:11 - rFactor2
    Frames: 8859 - Time: 113156ms - Avg: 78.290 - Min: 60 - Max: 122
    road with mirrors
    2017-05-17 22:34:35 - rFactor2
    Frames: 9233 - Time: 113703ms - Avg: 81.203 - Min: 61 - Max: 132

    cockpit with mirrors
    2017-05-17 22:36:41 - rFactor2
    Frames: 7807 - Time: 113250ms - Avg: 68.936 - Min: 52 - Max: 105
     
  5. Mulero

    Mulero Registered

    Joined:
    Jan 16, 2012
    Messages:
    351
    Likes Received:
    153
    @SPASKIS It is normal that with the cockpit camera there is less FPS. Since other cameras do not render the cockpit because it is not necessary, this allows a large increase of FPS.... probably. I also may be wrong... :)
     
    SPASKIS likes this.
  6. SPASKIS

    SPASKIS Registered

    Joined:
    Sep 7, 2011
    Messages:
    3,155
    Likes Received:
    1,420
    I can confirm the same issue regarding "trackside camera performance and its relationhip with having used cockpit view in the same outing" using latest beta and S397 DX11 ready content: USF2000@Mores.

    TRACKSIDE CAM
    2017-05-17 23:40:47 - rFactor2
    Frames: 3339 - Time: 49422ms - Avg: 67.561 - Min: 50 - Max: 84

    COCKPIT + MIRRORS
    2017-05-17 23:42:03 - rFactor2
    Frames: 3244 - Time: 49359ms - Avg: 65.723 - Min: 54 - Max: 88
    TRACKSIDE CAM
    2017-05-17 23:43:02 - rFactor2
    Frames: 2867 - Time: 49469ms - Avg: 57.955 - Min: 45 - Max: 70

    Other views do not show this behaviour

    HOOD + MIRRORS
    2017-05-17 23:51:32 - rFactor2
    Frames: 3383 - Time: 49437ms - Avg: 68.431 - Min: 56 - Max: 91

    COCKPIT + MIRRORS
    2017-05-17 23:52:36 - rFactor2
    Frames: 3223 - Time: 48828ms - Avg: 66.007 - Min: 56 - Max: 87
    HOOD + MIRRORS
    2017-05-17 23:53:37 - rFactor2
    Frames: 3360 - Time: 49360ms - Avg: 68.071 - Min: 58 - Max: 90

    ROAD + MIRRORS
    2017-05-17 23:54:42 - rFactor2
    Frames: 3630 - Time: 49391ms - Avg: 73.495 - Min: 61 - Max: 98

    COCKPIT + MIRRORS
    2017-05-17 23:55:39 - rFactor2
    Frames: 3262 - Time: 49219ms - Avg: 66.275 - Min: 56 - Max: 88

    ROAD + MIRRORS
    2017-05-17 23:56:39 - rFactor2
    Frames: 3613 - Time: 49062ms - Avg: 73.642 - Min: 62 - Max: 100


    You might be right but losing 15% fps because of a static image that could be refreshed at a much lower framerate is definitely something to look at. I would expect a fps increase considering the reduced % of screen that windshield covers.

    I could also say that trackside cameras need to render the outer part of the car from different angles and that it should be heavier than cockpit view which is not the case. In Monaco trackside performs 15%better than cockpit. Note that I never use opponents in benchmarks.

    In any case my point was different. Considering that cockpit view is somehow affecting my trackside performance (I wish other reported) I wonder if whatever causes it could be affecting the cockpit view performance itself. Considering the better performance other driving views it shouldn't be discarded to be caused by something different than the cockpit to be rendered.

    I understand that the lower increase between hood and cockpit with USF2000 is due to having to render the wheels. With Clios Hood view was even simpler, very close to road view. Road view increase above 10% fps. I guess it depends on the detail and complexity of the inside.
     
    Last edited: May 17, 2017
  7. juanchioooo

    juanchioooo Registered

    Joined:
    May 16, 2016
    Messages:
    1,205
    Likes Received:
    1,165
    Same I have explained wrong, the camera from the TV the viewer is lowered to 35 fps, the cockpit is going to 70 to 100 fps. regards

    Sorry i use the translator
     
  8. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    2,785
    Likes Received:
    1,378
    In light of Spaskis findings, it made me wonder if cockpit materials might have something to do with performance. My Palm Beach test above that shows improvements over DX9 used the fps-heavy Cobra, so what about the fps-light Corvette GT2, the car that least affects fps in cockpit view?

    settings: Level 4 AA, PPE=off, resolution 5760x1080 multiview, no sync, aggressive threads=off
    system: single GTX 1080, 8 GB VRAM; i7-4770 @3.4 GHz, 32 GB RAM
    circuit: Palm Beach
    car: Corvette GT2
    view: cockpit camera
    weather: overcast

    DX9, Avg 144 fps, Min 131 fps, Max 156 fps
    Build 1828309, Avg 131 fps (-9%), Min 118 fps (-10%), Max 144 fps (-8%)

    Clearly not the same improvement as the Cobra when compared to DX9!

    The difference between the Cobra results and the Corvette GT2 results are also far greater than my initial static cockpit tests in the Silverstone garage were, where the Corvette has only 8% more fps than the Cobra and here it's more like 20% difference when running the lap.
    ***
    The weather is improving here in Oregon and that means I will spend more time outside, so my benchmarking efforts will be skimpier for the upcoming week or two.
     

    Attached Files:

  9. FuzzyFassbender

    FuzzyFassbender Registered

    Joined:
    Apr 26, 2015
    Messages:
    100
    Likes Received:
    42
    There may be enough data collected already but thought I would throw in my unscientific observation. I'm getting approximately 20 to 30 (avg. 15%) "less" FPS with the DX11 beta. With everything being equal (settings wise) I took frame rate readings at various points around NOLA without AI. I sure hope DX9 performance will be reached as I just went back to triple screens recently based on DX9 performance. I can get very good triple screen performance in Assetto Corsa, but I would MUCH rather race rF2!

    5780x1080, fullscreen, multiview, AA=3, no PP for DX11
    Most in game display settings at Full except for reflections=Low, blur=Off, X8
    NOLA-C, sunny/clear, 10AM, private practice, USF2000, cockpit view

    B1825438
    i7 3770, GTX 1070, 16 GB ram
     
    Last edited: May 19, 2017
  10. Trevor Greenfield

    Trevor Greenfield Registered

    Joined:
    Apr 19, 2017
    Messages:
    37
    Likes Received:
    11
    Here's a full test for 5/19/2017 1832786
    Again, remember I'm using a base test config of Low/No everything and just 1 USF2000 car at silverstone with sunny conditions @ 1080p. You would never play like this, it looks worse than GPL. I'm doing it this way because these were two of the first items updated for DX11 and because it eliminates any other settings or AI issues.

    2017-05-19 09:57:42 - rFactor2 - NO PP
    Frames: 5981 - Time: 30000ms - Avg: 199.367 - Min: 181 - Max: 212
    • Standings/pre race massive FPS dip is gone. Still fluctuating FPS in pre-race menus but enough FPS that nobody should notice.
    • 6% gain in FPS Avg, 7% in FPS min, 0.5% drop in Max FPS
    • I've noticed a bug with 4x AF that it looks like it is not engaging. You can see in the numbers, but I have tried 3 times.
    • FPS is about to the level where you can play single screen 1080p at decent settings, but we're not quite to the level of DX9 yet (Most of these settings I didn't try yet in DX9 to be fair). We're still 14-20% off DX9.
    [​IMG]

    Let me know if you would like the XLS
     

    Attached Files:

    Last edited: May 19, 2017
    Louis likes this.
  11. F1lover

    F1lover Registered

    Joined:
    Oct 4, 2010
    Messages:
    349
    Likes Received:
    137
    4th world what ever your doing keep doing it man!!!! rF2 has never run soo good!!!!!
     
    demerzel likes this.
  12. BoothJoe

    BoothJoe Registered

    Joined:
    Oct 12, 2015
    Messages:
    202
    Likes Received:
    8
    Trevor, out of curiosity, what do you consider the minimum playable FPS level? I'm running in the 90s on triples with everything at high or full (USF2000 at NOLA), but no Post and I find this perfectly acceptable. I've moved my offline season to DX11 and don't see me going back to DX9.
     
  13. juanchioooo

    juanchioooo Registered

    Joined:
    May 16, 2016
    Messages:
    1,205
    Likes Received:
    1,165
    The graphics engine does not take advantage of dx11, much higher assetto
     
    Last edited: May 19, 2017
    kantenflimmern likes this.
  14. Trevor Greenfield

    Trevor Greenfield Registered

    Joined:
    Apr 19, 2017
    Messages:
    37
    Likes Received:
    11
    Hi Joe. Well believe it or not I moved to triples last year and I will never go back to a single screen unless its like a 4k 45"-60". But because even getting the triples above 60fps in race can be a tricky configuration, I'm more interested in matching the single 1080p FPS from DX9 first, so that at least I am starting from a similar base. 15-20% off will just make it that much more difficult to get triples above 60fps, and Ive verified this informally. After we get the single screen squared away then I will move back to triples.

    All that said, obviously the minimum is 60 fps @ 60 hz, and it has been that way for a long time. But the way in which the frames are displayed has something to do with it, and I'm seeing that DX11 is causing less stutter and tear at frames below 60fps visually compared to DX9. And the trick is, what is enabled/settings to get that 60+ fps. Like, AA1 getting me 60fps+ looks like a PS2 game, so that won't cut it. So there is a minimum standard of graphics that need to be achieved in 2017 at 60+fps in order for the game to look as good as whats available elsewhere.

    The ideal standard has been all over the board in terms of capping FPS satisfactorily, with some people reporting 60 fps, 61fps, 72fps, 90fps working for them and their systems, and depending on whether we're talking about rf2, AC, iracing or whatever other engine. Personally, I don't notice it being perfectly smooth in rf2 DX9 until I get to 120fps I can cap it at. At that rate, its as smooth as if I turn on vsync. But again, I'm noticing that less than 120fps cap in DX11 doesn't cause as much stutter or tearing disruption, so I might be able to live with something less.

    Vsync is usually something I turn to make this work - it just double buffers 60 fps and is silky smooth with far less GPU resources than trying to run 120fps. This also introduces display lag, which can be observed by many, and I know its there, I just can't see it myself.

    Within a couple of months, I will be moving to a new graphics card and end my woes, but right now I have a perfect opportunity to fight for the underpowered GPUs everywhere by testing and reporting ;)
     
  15. stonec

    stonec Member

    Joined:
    Jun 19, 2012
    Messages:
    3,253
    Likes Received:
    1,377
    AA levels between DX9 and DX11 are not comparable from the tests I did. Actually with AA X1 in DX11, certain things like white lines look better than AA X8 in DX9 (see my screens attached). There is little reason to use anything higher than AA X2 or X3 in the current DX11 build, the image is more or less identical.
     

    Attached Files:

    The Iron Wolf and Mulero like this.
  16. BoothJoe

    BoothJoe Registered

    Joined:
    Oct 12, 2015
    Messages:
    202
    Likes Received:
    8
    Here's my run from tonight. I7 4770 24Gram, Sapphire Rx480 8Gram, triples eyefinity only 5900x1080. Average FPS in the 80s. NOLA, USF2000. Tonight I ran at 9a.m. so that means long shadows. I have everything turned to High or Full or Max but no Post.
     

    Attached Files:

  17. Trevor Greenfield

    Trevor Greenfield Registered

    Joined:
    Apr 19, 2017
    Messages:
    37
    Likes Received:
    11
    For fun, I wanted to see if I can run triples.

    Here's my best BoothJoe config with everything on max except AA @ 2 with 1 USF2000 car @ NOLA, No PP:
    2017-05-19 19:20:31 - rFactor2
    Frames: 439 - Time: 30000ms - Avg: 14.633 - Min: 12 - Max: 19

    Now here's the bare minimum I would recommend for triples

    2017-05-19 19:30:36 - rFactor2 - Min @ NOLA Triples AA1
    Frames: 2676 - Time: 30000ms - Avg: 89.200 - Min: 76 - Max: 105

    This is everything set on off/minimum except player detail medium, texture detail full, shadows low, shadow blue optimal (so we're talking missing detail on tracks, no AF, No PP, basic shadows):

    2017-05-19 19:32:52 - rFactor2 - Min @ NOLA Triples AA2
    Frames: 1880 - Time: 30000ms - Avg: 62.667 - Min: 52 - Max: 76

    2017-05-19 19:34:36 - rFactor2 - Min @ Silverstone Triples AA2
    Frames: 2074 - Time: 30000ms - Avg: 69.133 - Min: 61 - Max: 81

    2017-05-19 19:36:10 - rFactor2 - Min @ Silverstone Triples AA1
    Frames: 2255 - Time: 30000ms - Avg: 75.167 - Min: 66 - Max: 88

    So with about another 15-20% (Back to DX9) I will be able to run triples with AA2 and 20 AI at some tracks without it going below 60fps. But right now there is no way I can add any detail, increase settings, or add AI without running out of fps.
     
  18. SPASKIS

    SPASKIS Registered

    Joined:
    Sep 7, 2011
    Messages:
    3,155
    Likes Received:
    1,420
    Build 1832786: Same performance as previous 1825438. The only difference seems that fps while in the garage do not drop to 14 fps anymore. As a summary: trackside cams are already up to DX9 level but cockpit view is around 20% below DX9.

    upload_2017-5-21_19-6-45.png
     
  19. Rony1984

    Rony1984 Registered

    Joined:
    May 10, 2012
    Messages:
    488
    Likes Received:
    23
    A little post on the input- lag that I experience with DX11.

    After the fullscreen fix there is less input-lag, but still notably more than with dx9. I've done a lot of back-to-back testing lately with different car/track combos and the result was always the same: very subtle yet notable more input lag.

    I shot 2 videos to capture the difference in input lag. Both with the car standing in the pits and having both the virtual and real wheel spokes in view as reference points, and me turning the wheel slowly and rapidly to have the virtual one trailing behind the real one. Unfortunately, the difference is so subtle that it doesn't show well on tape. Yet I feel that even on tape it's visible. The effect is more pronounced when actually driving, making the experience slightly like slow-motion. Again, the effect is subtle, but there. In the process of back-to-back testing, I've also invited several people to give their opinion on what they perceived, and all agreed that with dx11, the virtual wheel was trailing behind the real wheel more than with dx9, even the ones with absolutely no clue about computers and what they were looking at.

    One more (important?) thing: in the process, I've experimented with the maximum pre-rendered frames, both in dx9 and dx11, yielding a surprising result: in dx9, there was no difference between 'let the application decide', 4 and 1. Remarkably, there was no screen-tearing whatsoever with setting 1, which provided a super smooth at almost fully lag-free experience. With dx11 however, 'let the app decide', 4, 3 and 2 showed no difference. Setting 1 however, caused for terrible screen-tearing and a very stuttery experience(stuttery in addition to showing tears). I'm not savvy enough on subject to know what this might imply, other than that the native setting of dx11 is different from dx9?

    One last thing: I'm not an expert with video editing software and hence the 2 single files, sorry:)

    My system:
    CPU: Intel Core i7 3770 @ 3.40GHz, Ivy Bridge 22nm Technology
    RAM: 12,0GB Dual-Channel DDR3 @ 798MHz (11-11-11-28)
    Motherboard: MEDION H77H2-EM
    Graphics: NV Surround (5760x1080@60Hz), NVIDIA GeForce GTX 970 (MSI), multiview enabled, frames capped at 60

    DX9 vid: https://www.dropbox.com/s/vswammcrsmcbhkq/DX9.mp4?dl=0

    DX11 vid: https://www.dropbox.com/s/bm4j30mxqqw9cct/DX11.mp4?dl=0
     
  20. demerzel

    demerzel Registered

    Joined:
    Aug 30, 2015
    Messages:
    197
    Likes Received:
    127
    I'm not surprised about the pre-rendered frames. The fullscreen mode is still bugged. Sometimes I can still hear windows sounds when I press keyboard buttons even though the game is set to fullscreen. Which means it's still not running on proper fullscreen mode and windows still capture keyboard inputs. Also many times I have to click into the screen to actually control the game. The screen goes black for a second and than back to normal.
     
Thread Status:
Not open for further replies.

Share This Page