Thinking about improving FPS, a path to optimizing?

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

  1. dadaboomda

    dadaboomda Registered

    Joined:
    May 15, 2016
    Messages:
    316
    Likes Received:
    93
    There is a very very very strong lose of FPS when we use ac cobra compared to other ISI cars. Especially if environment reflections are turn ON.

    So you're right !

    Sorry I dont take the time to answer to one of your post concerning real driving VS rf2. I will try to do this when I could. But I don't think.my view is very pertinent because I have for my age a "strong" experience but I made this experience with a low cost car (low powerfull but exeptional engine fortunately, very very bad chassis + suspensions + frequently I bought used tire with the worst grip of the world etc... I deactivated ABS to train.)
    But the thing to train for me like I did was a exceptional training / experience (0 grip, good acceleration, very bad brakes etc.)
     
    Filip likes this.
  2. WiZPER

    WiZPER Member

    Joined:
    Oct 5, 2010
    Messages:
    1,438
    Likes Received:
    154
    You were refering to yourself as WE, in a quote of yourself... so does Smeagol from LOTR. I tend to asume that people have a sense of humour, maybe I should'nt.
     
  3. SPASKIS

    SPASKIS Registered

    Joined:
    Sep 7, 2011
    Messages:
    2,494
    Likes Received:
    791
    My experience is the same. The drop in frames depends on the viewing distance established in the track scn and on the car itself which might have more or less visible parts for that view.
     
  4. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    1,901
    Likes Received:
    521
    Since the McLaren generated the lightest fps load (106 fps) in your prior test, you probably should be looking at the other cars.
     
  5. Travis

    Travis Registered

    Joined:
    Dec 7, 2012
    Messages:
    65
    Likes Received:
    23
    I test with the default time of day...everything is largely default from the graphical settings to the track conditions. The only aspects I change is the amount of visable cars (20), AI (19) and the PP (medium).

    I'd love to read what improvements the testing team saw as I am not overly happy with the results; I see less than a 10% difference in performance and far less than what URD can manage with their mods with custom liveries from the rF2 community. It still amazes me that pCARS 2 runs like a dream at max settings with 31 AI cars and rF2 does not, even at default settings, with just 20 AI. I am very happy to read S397 will continue to work on performance as well as other aspects of rF2.

    Perhaps someone who is au fait with 3D car models can provide a detailed analysis of what the differences are between the S397 cars and the URD ones (as an example). I wonder if the interior reflections combined with the exterior reflections is cause for the frame rate issues as URD interiors look less reflective. Just a thought!

    Probably but I don't have much time to sit and test every car unfortunately. Yesterday was little more than a quick check of the updated performance.
     
  6. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    1,901
    Likes Received:
    521
    Believe that's called jumping to conclusions when you don't complete the testing.
     
  7. Travis

    Travis Registered

    Joined:
    Dec 7, 2012
    Messages:
    65
    Likes Received:
    23
    Thanks for the belittlement. I did state I recorded less than a 10% improvement across all GT3 cars.

    The fact is I shouldn't have to waste hours getting official content to run at default setting as a mod from URD does with NO fettling from max settings.

    Enjoy the update & all future updates.
     
    Last edited: Nov 3, 2017
  8. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    1,901
    Likes Received:
    521
    As I said in my original post, it comes down to polygons. The S397 McLaren is over 200k polygons (measured the AC version) vs. the URD Ferrucio which is in the 80k polygon range.
     
  9. Travis

    Travis Registered

    Joined:
    Dec 7, 2012
    Messages:
    65
    Likes Received:
    23
    Has Studio 397 confirmed the polygon count or is the number a bit of guess work due to AC which is stated to be 200K polygons.

    I do believe you may be onto something as the GPU load is much less on the ISI and URD made cars versus the S397 cars and the difference in minimum frame rates at default and then at max graphical settings for the GT3 Pack for me was less than 5 fps.
     
  10. stonec

    stonec Member

    Joined:
    Jun 19, 2012
    Messages:
    2,403
    Likes Received:
    600
    I'm pretty sure it's related to the car models. I would remove parts to see what is the problem if I could access these mods, too bad they are encrypted. The update has achieved something like a 5% performance improvement for me, but it's still less than the difference between McLaren and Radical FPS, McLaren achieves 15% higher performance. So IMO it's about the car model, not shadows.
     
  11. Travis

    Travis Registered

    Joined:
    Dec 7, 2012
    Messages:
    65
    Likes Received:
    23
    I concur with your findings :)

    The McLaren is by far the most optimised vehicle and offers up the best frame rates before and after the update. I recorded an extra 4fps with the McLaren using the default graphical settings. The Radical & Corvette remained the worst performing GT3 cars.

    I hope S397 find further more opitmisations soon as I'd love the official content to run as well as the mods I own.
     
  12. dadaboomda

    dadaboomda Registered

    Joined:
    May 15, 2016
    Messages:
    316
    Likes Received:
    93
    Agree, but at a moment we are just customers. We do it on our free time.

    For exemple :

    My ~50 hours of tests permise me to understand what 5-7 settings are the worst in terms of performance (I am not a gamer since 15 years, I was not an expert to do tests trying to find good graphics settings) :

    If just S397 have informed us via the forum with a comprehensive post or specific thread that the worst settings for performance are :

    shadows (worst), environment reflections, REAL mirrors, PP, number of AI of visible AI, quality of AI , AA (not specific to rf2 of course), acceleration time (for CPU put X1 or OFF) etc. (with more details)

    ---> Users would have had to do fewer tests, would have been less frustrated, would have had a much better image of rf2 and S397, would have been able to find more relevant and less random graphic settings while being free to take into account or not these informations, it would also drastically reduce the "complaints" that some people no longer seem to bear, etc.

    In short, the problems of S397 communication is not new.

    And above all essential:
    Users could easily wait for the graphics engine dx11 which is for the moment very bad in terms of performance (needless to hide it) that it is gradually improved (the performance of the shadows is the worst setting and is one of the real priorities of S397 = good news, but we know that since only 2 days).

    To count on the community to have help of this type is normal but to be able to count on the software publisher who knows well identified problems IT IS BETTER.
     
  13. dadaboomda

    dadaboomda Registered

    Joined:
    May 15, 2016
    Messages:
    316
    Likes Received:
    93
    Maybe for VR users it is important that the cockpit is "very" beautiful and very detailed: so I do not express myself for this type of users.

    For those who use 1 or 3 screens:

    I do not know you but I look at my cockpit not even 0.1% of the time while driving. And when I look at it is only the displays of the car, but that have a really very low interest.
    Moreover, and this is the most important element, the cockpit graphics have extremely little importance compared to the physics / FFB / interirior sounds of the car (this is my opinion): therefore improve the quality of the cockpit compared to cars ISI or URD has no interest.

    NUANCE: for VR users it may be that putting more detail in the cockpit is fine (I'm not talking about quality but details).
     
  14. stonec

    stonec Member

    Joined:
    Jun 19, 2012
    Messages:
    2,403
    Likes Received:
    600
    I applaud you for spending so much time to find the correct settings. I did several hours of testing in summer and came to conclusion that there is a PCI-E bottleneck with gMotor when using newer high-poly high-res cars and tracks. In other words, with PCI-E 2.0 bus a GTX 960 or GTX 1050 Ti is enough to get a PCI-E bottleneck. I would assume with PCI-E 3.0 you get the same bottleneck when using a GPU like GTX 1070/1080 and VR or multiview. Hence why people struggle in VR even with the best GPU money can buy.

    I think any optimization that S397 can do to car models is wellcome, but ultimately there is only so much that can be done with PCI-E limit as it is. I think if rF2 was native DX11 app, this PCI-E limit would be a non-issue. Other DX11 sims I tested use less than 10% of PCI-E 2.0.

    Here is a graph I made showing the bottleneck:

    rF2 scaling.png
     
  15. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    1,901
    Likes Received:
    521
    There's no need to confirm the polygon count of the McLaren with Studio 397 and I doubt they would tell us since the model is encrypted.

    The reason there's no need to confirm with Studio 397 is the AC template and the rF2 template show the same polygons for the exterior of the McLaren. Because there are no templates for the interior, then examining screenshots of the interior shows that the interior detail is the same by looking for the vertexes. Same shape of air vents, same shape of LCD cover, etc. Thus the only difference, if any, between the AC and rF2 versions of the McLaren are going to be the wheels/tires at the LOD A level.

    Getting the polygon count for unencrypted models is merely opening the model inside 3DSimEd and reading the report.
     
    Last edited: Nov 3, 2017
  16. Emery

    Emery Registered

    Joined:
    Oct 24, 2010
    Messages:
    1,901
    Likes Received:
    521
    Probably worth noting that polygon counts for LOD B, C, D aren't known for the Studio 397 cars. They may or may not have used the same models as Kunos. For what it's worth, the Kunos numbers (rounded up) are 28k, 7k, and 3k polygons respectively.

    What becomes obvious by looking at polygon count for the LODs is that being aggressive on LOD switching has rewards for fps performance. You can have 8 cars visible at Kunos' LOD B level for the fps price of 1 car visible at LOD A level. On the grid, you might be looking at 4 to 8 cars as LOD A level (your own row plus the next row or two or three), thus big fps impact.
     
    Last edited: Nov 3, 2017
    Ari Antero likes this.
  17. WiZPER

    WiZPER Member

    Joined:
    Oct 5, 2010
    Messages:
    1,438
    Likes Received:
    154
    Looking at the S397 GT3 McLaren vs. the URD C7R, the MAS-size difference alone is 89Mb vs 41Mb (textures plus GMTs). Then add the fact that the new GT3s are using different body shaders, features dynamic damage and possibly made ready for visible rain-effects.
     
  18. LokiD

    LokiD Registered

    Joined:
    May 3, 2017
    Messages:
    298
    Likes Received:
    191
    how is the GT3 pack running on a 1080? anyone have one.. Im still hearing fps issues with this pack, and its really really putting me off getting it as i still (yes mentioned again) have drops with the radical.. Unless you run around 3 ai lol!
     
  19. WiZPER

    WiZPER Member

    Joined:
    Oct 5, 2010
    Messages:
    1,438
    Likes Received:
    154
    Bollocks, I can run 40 AIs @ 90fps midfield on the grid.
     
  20. stonec

    stonec Member

    Joined:
    Jun 19, 2012
    Messages:
    2,403
    Likes Received:
    600
    With shadows at what setting?
     

Share This Page