1. Do you need support for Assetto Corsa Competizione? Please use the proper forum below and ALWAYS zip and attach the WHOLE "Logs" folder in your c:\users\*youruser*\AppData\Local\AC2\Saved. The "AppData" folder is hidden by default, check "Hidden items" in your Windows view properties. If you report a crash, ALWAYS zip and attach the WHOLE "Crashes" folder in the same directory.
  2. FOR ASSETTO CORSA COMPETIZIONE: If you report issues with saved games, please always zip and attach your entire User/Documents/Assetto Corsa Competizione/Savegame folder!
  3. If your game executable is missing, please add your entire Steam directory to the exceptions in your antivirus software, then run a Steam integrity check or reinstall the game altogether.

The ACC CPU Benchmark Thread

Discussion in 'ACC General Discussions' started by Deatroy, Dec 7, 2018.

  1. liakjim

    liakjim Alien

    Clear and day conditions are enough. The 480 was at 60-70% so i think it's enough.
     
    Deatroy, LeDude83 and After_Midnight like this.
  2. MR279321

    MR279321 Rookie

    Here are my results.

    My PC
    Intel 7700K running at 4,8Ghz
    Geforce 1080 +200Mgz overclocked
    32gb ram


    I did 3 tests with with 1080p, 2k and 4K resulution.
    I also addet 3 videos were you can see GPU temperature GPU Load and GPU Clock speed.
    In my opinion it is very important to log GPU and CPU load to find out if we have a CPU or GPU Bottleneck.

    What i found out after my tests is playing at 1080P and even 2K resulution with lowest grafic settings my 7700K is bottlenecking.
    In the first 2 videos you can see that GPU Load is very low and jumping up and down. This means my GPU is waiting for my CPU
    until the calculations of the CPU are done.

    In the last video with 4K resulution you can see my GPU is nearly at 100% GPU Load.
    This means the CPU Bottleneck is gone and my GPU is runing at its full power.

    1080P Test
    2018-12-08 19:28:17 - AC2-Win64-Shipping
    Frames: 7563 - Time: 60000ms - Avg: 126.050 - Min: 98 - Max: 149


    2K Test
    2018-12-08 18:43:19 - AC2-Win64-Shipping
    Frames: 7804 - Time: 60000ms - Avg: 130.067 - Min: 105 - Max: 151


    4K Test
    2018-12-08 18:56:44 - AC2-Win64-Shipping
    Frames: 7438 - Time: 60000ms - Avg: 123.967 - Min: 109 - Max: 143


    As you can see there isn't much difference between the average Framerate from the 1080P and the 4K test
    The reason is at 1080P my GPU Load was very low an my GPU wasn't nowere near its performance limits.
    Remember playing at 4K your GPU needs to push 4 times more pixels than in 1080P an my framerates are nearly identical because my GPU is running 100% now.
     
  3. baakstaff

    baakstaff Racer

    Definitely good information to have; it'd also be useful to see if any single core is at 100% along with overall cpu usage.



    For example, in this video using the benchmark settings, the top line is all gpu info; if the core clock hits ~1570 mhz, then its gpu bound. Below that there's two pieces of cpu info: the first number is the highest % usage on any single thread, showing how hard a single core is being hit. The second is my overall cpu usage, the average of all threads. The bottom row is my framerate. I never saw higher than 73% usage on any single core.

    I used Relive to record, so the only thing sapping cpu performance would have been hwinfo64 with a polling rate of 500 ms.

    My results according to fraps end up like this (seperate run without hwinfo64 or relive running):

    2018-12-08 14:35:51 - AC2-Win64-Shipping
    Frames: 5312 - Time: 60000ms - Avg: 88.533 - Min: 72 - Max: 105
     
  4. luxxx797

    luxxx797 Simracer

    here we go, settings as in first post:

    2018-12-08 21:43:05 - AC2-Win64-Shipping
    Frames: 6402 - Time: 60000ms - Avg: 106.700 - Min: 86 - Max: 144


    and settings maxed out:

    2018-11-25 12:07:31 - AC2-Win64-Shipping
    Frames: 9969 - Time: 60000ms - Avg: 166.150 - Min: 132 - Max: 201

    ?? how is possible to get 60% higher fps with all at max????
     
    Last edited: Dec 9, 2018
    LeDude83 likes this.
  5. BrunUK

    BrunUK Alien

    Since I've posted before about the effects of opponent visibility on FPS, I thought it would be a good idea to take another look at that with this test setup. Besides the difference between setting it to ALL/1, it's also interesting to note the effect of letting the field drive out of view before benchmarking. Here's the results...

    Code:
    A: Opponent visibility 'ALL', driving in pack:    Avg: 116.567 - Min: 097 - Max: 135
    B: Opponent visibility '1', driving in pack:      Avg: 235.900 - Min: 192 - Max: 289
    C: Opponent visibility 'ALL', driving solo:       Avg: 145.183 - Min: 126 - Max: 154
    D: Opponent visibility '1', driving solo:         Avg: 260.917 - Min: 202 - Max: 339
    You can immediately see by comparing A/B that only rendering one other car causes a huge increase in framerate, even though the rest of the field is close by.

    Comparing C/D, you would expect C to increase towards D as fewer cars are rendered but it only improves slightly compared to being in the middle of the pack, even when literally no other cars are actually visible. There seems to be something very wrong with ACC if opponents that are potentially visible have such a crippling effect on performance even when they're not on screen.

    The important thing to note about this is that there's no difference to the physics and AI calculations. Each scenario has a full grid of 20 opponents. There's always been an assumption that ACC is heavy on the CPU because of physics and AI. This seems to suggest that's not the case at all, and it's actually rendering the cars which is causing poor performance.
     
  6. baakstaff

    baakstaff Racer

    One thing i noticed when testing this was that at night in the wet you could still see reflections on the road of the taillights of the cars that weren't being rendered, and that their position was being updated at a significantly lower rate than the visible car, to the point that they were visibly stuttering along the road. So I'd guess that there is some shortcut being taken when performing the calculations of the other cars' position and speed.

    Were you GPU bound with opponent visibility set to "All"?
     
  7. liakjim

    liakjim Alien

    Something strange is going on.

    Στάλθηκε από το m2 note μου χρησιμοποιώντας Tapatalk
     
  8. luxxx797

    luxxx797 Simracer

    mistery solved....the results at max settings was from the GPU benchmark test, that was in hotlap mode.
    sorry for the mistake
     
  9. GRFOCO

    GRFOCO Alien

    wow, that's very surprising, in a negative way.
     
  10. LeDude83

    LeDude83 Alien

    So it appears that stuff is being rendered and takes a performance toll although it's not visible on the screen, correct?

    Sounds like something that might be fixed easily. Might also be very interesting for taxing trackside objects.
    Seems like there could be quite some improvement potential.
     
    Deatroy likes this.
  11. Deatroy

    Deatroy Alien

    Oh nice what i can see here.
    Should we move to 720p and without rain?

    @BrunUK seems that rain causes more taxing on the GPU.
    We see AMD User hang in GPU Limit with lowest graphic level.
    I noticed it aswell with my R9 285.
     
    liakjim likes this.
  12. LeDude83

    LeDude83 Alien

    It's your test method so your call. Instead of dropping the res though I'd reduce visible cars.
    Very good thread btw. Kind of inevitable to have 1 or 2 method adjustments on the way. CPU is kind of tricky to isolate for everyone.
     
  13. BrunUK

    BrunUK Alien

    No GPU limits in any of the tests. With all opponents visible it was generally only around 40%, compared to 60% with minimum opponent visibility. The latter being higher because of the huge increase in framerate despite having only one other car to render.

    For the hell of it, I repeated the test with all graphics options at 'Epic' and it's still CPU limited at 1080p, with GPU use up to 75% (Avg: 59.517 - Min: 48 - Max: 75). Have to go up to 2560x1440 before I see 99% GPU usage, and even then there are drops to 90%.
     
    baakstaff and LeDude83 like this.
  14. CPU: i7 5820k 4,1 GHz
    GPU: GTX 970
    RAM 16GB 2400 MHz

    2018-12-09 17:36:37 - AC2-Win64-Shipping
    Frames: 6068 - Time: 60000ms - Avg: 101.133 - Min: 86 - Max: 125

    It seem ACC doesn't take advantage of more than 4 cores. :(
     
    Last edited: Dec 9, 2018
  15. baakstaff

    baakstaff Racer

    Interesting, the lower gpu usage with all visible opponents could mean that we're still cpu bound because either the calculations for the cars positions are being updated much more frequently, or there's more work going into the physics calculations for the AI, more resources may be getting loaded in, or something else along those lines. Either way, it's not just the gpu thats having to work harder with all cars visible.
     
  16. Deatroy

    Deatroy Alien

    Interesting is that 7700k 4,8Ghz is better than 8700k 4,9Ghz and 9900k 5ghz.
    Like Chris said, UE4 seems to take no advantage of more cores.

    I will switch the testing in the week.
    No Rain, 720p to allow AMD Graphic Users to collect usefull data.
    You can start already benching guys if you want. Will update table later on.
     
  17. Deatroy

    Deatroy Alien

    Intel has ringbus and not a mesh. Due that the lantency's are low as wanted. Games love low lantency, due that Skylake-X is slower out of the box.
    Ryzen IF is comparable to Skylake-X Mesh, thats why those two should optimise RAM and not coreclock.

    Table is updated soon. Just sum up all results.
     
  18. Daniel Costa

    Daniel Costa Racer

    You need to keep some settings at EPIC to fully load the CPU.

    Mirror, View distance, shadows, foliage, needs to be high quality.

    I also recommend limiting the benchmark to the formation lap, where conditions are the same for everybody (assuming you start always from the same position). The car should also be starting from around the middle of the grid, to put more load on the mirror geometry rendering (more cpu demand).

    Last but not least, it should be a replay run, not a live race, so we can remove AI processing from the equation. AI should be a separate bench.

    edit: 720P if possible, to make sure.
     
    Last edited: Dec 11, 2018
  19. LeDude83

    LeDude83 Alien

    The claim on view distance, shadows and foliage needs test runs for verification/falsification. Can you deliver?

    The recommendation of replay mode I don't understand. AI processing = CPU load so clearly what is to be tested here.
     
  20. liakjim

    liakjim Alien

    I don't agree

    Στάλθηκε από το m2 note μου χρησιμοποιώντας Tapatalk
     
    luxxx797 and LeDude83 like this.
Similar Threads
Forum Title Date
ACC General Discussions ACC - 0.5.2 - Performance Review by GameGPU (29 GPUS & 27 CPUS) Jan 22, 2019
ACC Troubleshooting gamebreaking stuttering after massive ACC crash CPU Oct 14, 2018
ACC Troubleshooting I have CPU Thermal Throttling and BSOD in ACC Oct 12, 2018
Chit Chat Room AC/ACC & CPU's / Threads & Cores Sep 22, 2018
ACC Graphics - Tracks and Cars ACC - CPU Bottlenecking Sep 22, 2018
ACC Online Leagues / Championships room New ACC Porsche GT3 Cup Championship Wednesday at 10:23 PM
ACC Online Leagues / Championships room ACC support in RaceApp: Soon! Tuesday at 6:01 PM
ACC Controls & Peripherals What in game Accuforce DDv2 wheel settings are people using? Monday at 11:46 PM
ACC Controls & Peripherals ACC - Rotation whell degrees Problems Monday at 3:07 AM
ACC Multiplayer Unable to change server name; resets to "ACC Server (please edit settings.json)" Sunday at 11:02 AM
ACC General Discussions Dear Kunos!Kindly provide steering range information for each car in ACC. Aug 10, 2019
ACC Multiplayer For server admins: ACCServer 1.0.8 preview 1 Aug 9, 2019
ACC Online Leagues / Championships room Octane Online Racing - ACC Series (Oceania & Asia) Aug 9, 2019
ACC General Discussions '27 Display for ACC and AC Aug 9, 2019
ACC Virtual Reality Valve Index, SteamVR beta, ACC 1.0.7 beta Aug 8, 2019

Share This Page

  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice