Perfect SLI Scaling w/ rFactor 2 - Great Job ISI!

Discussion in 'General Discussion' started by Spinelli, Jul 12, 2015.

  1. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    31

    Settings and PC Info


    - ISI NSX and Portugal, 12:00 AM, rain, no opponents, pit garage
    - GPU: 2x GTX 970 (SLI)
    - Nvidia Driver: v353.30
    - Nvidia Profile: All default except using "prefer maximum performance" and max pre-rendered frames @ 1.

    NOTE: The overall framerates for these tests aren't nearly as high as could be due to the tests being performed while my GPU, RAM, and CPU were at the following slow default speeds:
    - GPU @ 1315/7000 MHz core/memory (instead of 1500/7500 MHz)
    - CPU @ 3.6 GHz (instead of 4.5 GHz)
    - RAM @ 16-16-16-39 2T timings (instead of 11-11-11-26 1T)


    TEST #1



    [​IMG]

    [​IMG]



    Default SLI Compatibility Bits - 0x02402005 (Sanctum, UAZ Racing 4x4, rFactor 2[rFactor2 Mod Mode.exe], etc.)

    With the default SLI comp. bits I had some real weird results...
    - Cockpit camera, both GPUs around 55% usage and poor SLI scaling and framerates
    - TV cam, same thing, both GPUs around 55% usage and around 75 fps
    - Chase cam (or any other cam like the cam near the wheels, side panel, etc.) and the framerates rise to 145 fps with both GPUs showing 98% usage. Wow!
    - Then I go back to TV cam and, unlike the first time, the GPU usage stays at 98% and framerates in the 140s

    Basically cockpit cam always SLI scaled poorly, any other cam SLI scaled great, and TV cam SLI scaled poor if the previous cam was cockpit but SLI scaled great if the previous cam was a cam that SLI scales great (e.g. chase cam)...Weird.



    Using SLI Compatibility Bits: 0x02D04005

    I just tried the SLI compatibility bits that I've seen others mention as of RF2 b982 called "0x02D04005 (LEGO Batman 2, LEGO Pirates of the Caribbean: The Video Game, LEGO Lord of the Rings, LEGO Star Wars III: The Clone Wars, LEGO Harry Potter: Years 1-4, LEGO Star Wars Saga, LEGO Harry Potter: Years 5-7)" and I'm happy to report that all issues are now gone and the GPU usage is at 98% for both GPUs and no matter which view is used, including the cockpit view where I'm getting over 130 fps :)


    [HR][/HR]

    TEST #2 - Both Reflections @ "High"
    - everything else same as above



    Default SLI Compatibility Bits

    - GPU Usage = 55% (both GPUs)
    - Framerate = 60 fps

    Note: Always poor SLI scaling no matter the view


    SLI Compatibility Bits 0x02D04005 (LEGO Batman 2, etc.)


    - GPU usage = 98% (both GPUs)
    - Framerate = 120 fps

    Note: Wow! Each GPU's usage stays in the mid-to-high 90s % no matter what view is used and framerates are doubled. FANTASTIC!


    [HR][/HR]

    TEST #3 - Max Graphics -
    except FXAA (disabled) and AA (Lvl 2)



    Default Compatibility SLI Bits

    - GPU Usage = 55% (both GPUs)
    - Framerate = 42 fps

    Note: Always poor SLI scaling no matter the view, not only that but there was major graphics corruption (solid coloured pink stuff) on the outer monitors.



    SLI Compatibility Bits 0x02D04005 (LEGO Batman 2, etc.)

    - GPU Usage = 98% (both GPUs)
    - Framerate = 88 fps

    Note: Even with all graphics settings maxed-out, SLI scaling remains pretty-much perfect (and no matter the view). FANTASTIC!




    TEST#4 HERE (different car and track combo)



    Conclusion: Until Nvidia sets it as default for rFactor 2, try SLI Compatibility Bits -
    0x02D04005 (LEGO Batman 2, etc.). The SLI experience, for me, goes from "broken" to perfect!
     
    Last edited by a moderator: Jul 17, 2015
  2. dsuspense

    dsuspense Registered

    Joined:
    Jan 18, 2013
    Messages:
    107
    Likes Received:
    0
    Nice analysis and writeup.
    I was always wondering why rF2 was the worst performer compared to GSCE, AC, and pCARS.
    These latest SLI compatibility bits really bring out the true performance of the rF2 gfx engine.
     
  3. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    31
    I just tried the same compatibility bits with SCE


    Default compatibility bits (0x00000000): ------ Avg: 202.036 - Min: 151 - Max: 265
    SLI Bits 0x02D04005 (LEGO Batman 2, etc.): Avg: 310.336 - Min: 229 - Max: 406

    :)

    The weird thing w/ SCE is that GPU usage was the same in both tests - high 80s to low 90s % for both cards. It seems that GPU usage shouldn't be the only thing looked at when determining SLI GPU scaling.
     
    Last edited by a moderator: Jul 13, 2015
  4. TechAde

    TechAde Registered

    Joined:
    Oct 13, 2010
    Messages:
    605
    Likes Received:
    36
    I'm noticing an issue when mirrors are turned off when using 0x02D04005 in that the mirrors flicker rapidly from black to grey (if mirrors have not been turned on during the session) or grey/black to the last image that was shown in the mirror (if mirrors have been enabled at some point during the session). If mirrors are enabled there is no problem.

    Any one else seeing the same?
     
  5. Nazirull Safry Paijo

    Nazirull Safry Paijo Registered

    Joined:
    May 28, 2011
    Messages:
    1,417
    Likes Received:
    35
    Nice work Spinelli
     
  6. Satangoss

    Satangoss Registered

    Joined:
    Jun 2, 2011
    Messages:
    1,123
    Likes Received:
    7
    Oh no... Not this again. It's well proven each computer reacts totally different under same nvidia parameters. For example, this "magical" bits does nothing to me (actually it gets the things worse). The statement about different SLI scaling depending on camera is simply non sense. Be happy for your good performance but don't make propaganda about the fictitious "amazing SLI scale" because it's just not true for a huge member of users whatever bits they use. BTW, the SLI properties in the bottom of Inspector bypasses the very bit you're using.
     
  7. dsuspense

    dsuspense Registered

    Joined:
    Jan 18, 2013
    Messages:
    107
    Likes Received:
    0
    GSCE tests:

    Track - Spielberg Vintage
    Car - Formula Retro Copersucar

    Default compatibility bits (0x00000000): Min: 74 - Max: 94
    SLI Bits 0x02D04005 (LEGO Batman 2, etc.): Min: 124 - Max: 150
    SLI Force alternate frame Rendering 2: Min: 124 - Max: 150

    Seems SLI bits and SLI AFR2 have the same effect for GSCE on my system.
     
  8. Flatspotter

    Flatspotter Registered

    Joined:
    Oct 7, 2010
    Messages:
    325
    Likes Received:
    77
    So cockpit cam still gives poor performance no matter what? That's a deal-breaker.
     
  9. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    31
    Did you try the exact dame test as me?

    It's not non-sense. I tested it. Why would I lie? I repeated the tests numerous times with the same results every time. The only thing that's non-sense is you and your aggressive post. What is your problem?

    So noone can change SLI bits? It gets bypassed? Really? So my two completely different SLI results, GPU usage results, and framerates are just a figment of my imagination? All the different SLI results thousands of players (including RF2 players) have had over many years using different SLI bits are all figments of our imaginations? Seriously, just stop.


    You're welcome for trying to share my results and trying to help people.

    Incorrect. Cockpit cam, and any cam, works brilliantly with SLI when using the correct SLI bits as my tests clearly show.
     
    Last edited by a moderator: Jul 14, 2015
  10. WhiteShadow

    WhiteShadow Registered

    Joined:
    Feb 16, 2015
    Messages:
    681
    Likes Received:
    3
    Are you saying that you get 120fps at Portugal with 30 cars at start with HDR, multiview, shadows, road-and environment reflections ?
     
  11. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    31
    I should have made clear that the test was not a race start or with other cars. It was in the pits, alone. Having said that, I used nighttime (including headlights) and rain due to them being much more graphically demanding (especially when used simultaneously) than daytime and no rain.

    As for the other gfx settings, they're noted in my OP.
     
    Last edited by a moderator: Jul 14, 2015
  12. Lazza

    Lazza Registered

    Joined:
    Oct 5, 2010
    Messages:
    8,782
    Likes Received:
    3,985
    The thread is called "SLI scaling". Cars weren't mentioned. Environment reflections are shown as off in the screenshot above. Your post suggests all you saw were "Portugal", "120fps" and "works brilliantly".

    A thread provides a context, it helps if you stay in it.
     
  13. Frank1463

    Frank1463 Registered

    Joined:
    May 18, 2014
    Messages:
    2
    Likes Received:
    0
    Thanks once again Spinelli for the usefull information.
    I tried the lego batman bits and got exactly the same results as I get with the crysis 3 bits (which sounds similar to your results. I have 2 x GTX 780's) so I wonder if it could depend on which nvidia driver we use. Recently I was using the latest 353.38 drivers and I was better off with just the 1 card. I tried the crysis 3 bits and the F1 2010 bits which have both worked well in the past but not with 353.38. I went back to 347.88 and now crysis 3 and lego batman both work well.
     
  14. WhiteShadow

    WhiteShadow Registered

    Joined:
    Feb 16, 2015
    Messages:
    681
    Likes Received:
    3
    The thread is called "SLI scaling", stay in it smart guy.
     
  15. Nuno Lourenço

    Nuno Lourenço Registered

    Joined:
    Oct 19, 2010
    Messages:
    592
    Likes Received:
    63
    Guys you need to understand that SLI or crossfire its like gambling. You can buy two systems with the same hardware and software and you have the risk that work in one and not in other... There's no 2 equal hardware components...
     
  16. Noel Hibbard

    Noel Hibbard Registered

    Joined:
    Oct 5, 2010
    Messages:
    2,744
    Likes Received:
    29
    I think the point of this thread is to point out that SLI is finally "scaling" well. To be precise, two cards are getting better framerates than a single card. I'm sure if you throw in more cars the FPS will come down, possibly even below 120 but it's still scaling. Although throwing in a bunch of AI puts more load on the CPU so you may end up not scaling as much because you will no longer be GPU bound but instead CPU bound. But if you only race online then AI is never even used.

    Satangoss seems to be getting different results though. From what I am reading, more people are seeing good scaling on this build with the Lego bits vs not. It think to know for sure what is going on it's best to stick to Apples and Apples when comparing results. So same cars, same track, same driver version, same bits and same settings. Personally if we are simply looking at scaling (which is the topic of this thread) then AI should be left out.
     
  17. DurgeDriven

    DurgeDriven Banned

    Joined:
    Mar 20, 2012
    Messages:
    6,320
    Likes Received:
    42
    So it does what most games already do. ?


    Great..... :rolleyes: >>>> hehehhe
     
  18. WhiteShadow

    WhiteShadow Registered

    Joined:
    Feb 16, 2015
    Messages:
    681
    Likes Received:
    3
    You are using Nvidia Surround bezel correction "subviews=1" which gives you much higher FPS then build 982 "subviews=2" added multiview adjustments.
    Your settings "subviews=1" is +120fps, "subviews=2" is 70fps, GPU loads are the same with both settings.

    Conclusion: If you are not rFactor veteran and don`t know all the tricks in the book the build 982 multiview ("subviews=2" added multiview adjustments) SLI experience is still pretty-much "broken".
     
  19. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    31
    Glad to hear it's working for you too, Frank.

    I'm going to do more tests including an identical test to the OP but with my CPU and RAM overclocked. I was running them @ default speeds rather than my "standard" overclocked speeds.

    Don't forget, everyone, even though there weren't any AI, the combination of night time and rain is quite graphically demanding compared to day time and no rain.

    Oh, I see, I see. Yes, as you pointed out, I didn't test the new style of multiview. I only tested the original style since that's the one we've had for years and hardly ever worked for anyone when it came to SLI.

    Having said that, I'm quite surprised the new style multiview has such a dramatic difference (according to your tests), or any difference at all for that matter, to the old style because - theoretically speaking in-terms of rendering - they shouldn't be any different from eachother other than a pre-set and locked camera angle and FOV VS an adjustable one. Weird.
     
    Last edited by a moderator: Jul 14, 2015
  20. Noel Hibbard

    Noel Hibbard Registered

    Joined:
    Oct 5, 2010
    Messages:
    2,744
    Likes Received:
    29
    SubViews=1 isn't even a valid option. It's either SubViews=0 (meaning there aren't any sub views) or SubViews=2 (meaning there is a main view in the center, a sub view on the left and a sub view on the right). It's been this way all the way back to rF1. SubViews=0 just means it scales a flat image across 3 monitors. SubViews=2 means it puts the normal straight ahead angle on the center monitor and a left angle on the left monitor and a right angle on the right monitor. So essentially the game is rendering three scenes. It's going to add a lot of extra load.

    But Spinelli, in your screenshots in the first post I see a checkmark on "MULTI VIEW" so your tests were with SubViews=2. The only thing added to this new build with regards to SubViews was the additional settings for the side angles and the bezel correction (But only if UseSubViewParams=1 which is the default). I don't think the new settings impact the performance though. But one could do back to back tests with UseSubViewParams=1 and UseSubViewParams=0.

    BTW, if you manually edit the config.ini and set SubViews=1 the game will actually ignore it and fall back to SubViews=0. That is why SubViews=1 performs and looks just like SubViews=0.
     

Share This Page