High DX11 CPU overhead, very low performance.

Discussion in 'Videocards - AMD Radeon Drivers Section' started by PrMinisterGR, May 4, 2015.

  1. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    Then funny how my scores can differ so much and overclockgn makes it worse
     
  2. Romulus_ut3

    Romulus_ut3 Master Guru

    Messages:
    780
    Likes Received:
    252
    GPU:
    NITRO+ RX5700 XT 8G
    To me, the R9 3XX-Fiji drivers seem to be the most consistent.
     
  3. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    Have you checked for frequency fluctuations in the GPU? Also if you have an AMD GPU and you OC its memory, it will clock down instead of artifacting.

    Furthermore, make sure that in the Windows power plan you use, the "PCI-Express power saving" option is on OFF.
     
  4. Bigbeard1986

    Bigbeard1986 Guest

    Messages:
    101
    Likes Received:
    3
    GPU:
    16gb
    Clock down? Mine will artifact, or worse, hardlock my system.

    If i go to msi afterburner right now and raise the memory OC very high, my system will black screen. That doesn't happen to you?
     

  5. Dygaza

    Dygaza Guest

    Messages:
    536
    Likes Received:
    0
    GPU:
    Vega 64 Liquid
    Try using 3dmark with only 3 cores active (3c/3t, no hyperthreading), You should get a lot less variation when windows scheduler don't mess with core usages. You will be cpu limited with mantle around 12M though, so don't compare that.
     
  6. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    Like I said I force gpu clock to run at full speed all the time when doing 3dmark test. Also my cpu is running at 4.6ghz regardless of load. so nothing should clock down. I do test about 3 times each time and results only vary slightly so I see it as accurate to my system. 4.6ghz really does improve my performance a bit as I get less stutter and a little bit better fps in some games.

    Anyway I find it strange. Could be my antivirus doing some sort of scan but it's set to do scan only when pc is idling. I'm using kaspersky if that helps. Anoter thing is i always do a clean boot before testing
     
  7. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    I only have Windows Defender and Malwarebytes, and they are both used for on-demand scanning when something is downloaded, never as resident.
     
  8. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    ok so i did a clean boot again and 3 runs on 720p default settings

    run 1

    dx11st = 1 497 264
    dx11mt = 1384 983
    mantle = 17 336 440

    Run 2

    dx11st = 1 445 210
    dx11mt = 1 326 386
    mantle = 17 146 873

    run 3

    dx11st = 1 420 148
    dx11mt = 1 297 649
    mantle = 16 843 085

    I only had the overhead test running with radeon pro set everything to run at max clocks. Cpu was also running 4.6GHz all the time. Funny how my performance degrades over time. I doubt it's a overheating issue as my cpu never peaks 64C and gpu never peaks over 63C when doing tests. Fans also spins at 100% so no way vrm can overheat.


    I also disabled ULPS. Funny when testing dx11 i see gpu usage fluctuate between cards. my gpu usag is sometimes as follows.

    GPU1 = 15%
    GPU2 = 8%

    On mantle gpu usage is exactly the same on both gpu's

    Just ran it again and vrm temps according to GPU-z stays under 50C
     
    Last edited: May 31, 2015
  9. Romulus_ut3

    Romulus_ut3 Master Guru

    Messages:
    780
    Likes Received:
    252
    GPU:
    NITRO+ RX5700 XT 8G
    I have done 5 runs for each set of drivers, used via the .dll injection method. Specs are on my profile. You can now clearly see the difference in Windows 8.1 Official WHQL and Beta drivers versus the Windows 10 drivers obtained via Windows update, modded to work on Windows 8.1.

    [​IMG]
     
    Last edited: May 31, 2015
  10. Dygaza

    Dygaza Guest

    Messages:
    536
    Likes Received:
    0
    GPU:
    Vega 64 Liquid
    Curious where that difference between 1023.3 and 1036 comes, they're both using same d3d driver version (.01128)
     

  11. Romulus_ut3

    Romulus_ut3 Master Guru

    Messages:
    780
    Likes Received:
    252
    GPU:
    NITRO+ RX5700 XT 8G
    The Catalyst 14.12 aka Omega and Catalyst 15.3 had the same D3D driver version too, but that doesn't mean they are the same. I assume that's the case with 1023.3 and 1036.
     
  12. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    Ok I don't know where to post but I've been playing around with the graphics settings and so on in bf4 and to test smoothness. I must say when I play on 95fps limit and vsync on my 96hz panel and vs mixture of ultra and high and low I can manage on both 95fps all the time using mantle. but on low I feel it's so much smoother its insane. also when I disabled crossfire I can only tell you i can't wait for the day to get back to single gpu as the smoothness is just insane. So much better than crossfire and now it feels so laggy on crossfire. +-5ms-10ms frametime vs 8ms - 15ms for the gpu suing the built in graph. I even play better and a lot easier to aim.
     
  13. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    I have similar results, where I feel that the performance somehow degrates over time, but then I run a 3rd or a 4th time and it seems ok again. I have no idea if it is a problem of the test or the hardware somehow.


    There are two things that might have happened. They either didn't bump the version but they changed something, or they changed compiler options for the driver libraries. Or maybe even both.

    The difference between the two branches of the driver is insane. Funny how 1018.1 is still (marginally) the fastest driver in this test.

    With which drivers you get crossfire judder in bf4?
     
  14. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    all of them and it's due to multigpu. It's because the frametime is much lower with single gpu and will be that way for a long time unless they use SFR
     
  15. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    And we can't even blame AMD for that anymore too :p

    I was under the impression that multigpu under Mantle was generally good.
     

  16. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    well it's good if they use sfr. And bf4 mantle implementation is just bad. in thief for example it's very good. It's very smooth with no stutter.


    Yeah we can't blame amd for high multigpu frametime cause i bet nvidia also have same issue. Might be less than amd but still the reason why people tend to suggest going with the best single gpu.
     
  17. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    What about DX11?
     
  18. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    well on dx11 I get some good fps. The smoothness is better with mantle when using crossfire but using dx11 I get better fps when running on single gpu in certain areas. so it's a mix bag.

    Maybe the wddm 2.0 will reduce memory usage when using mantle and give me a more stable fps
     
  19. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    When Windows 10 are out, I want to compare the last driver of the old branch on Windows 8.1, to the last of the newer branch, and both to the same on Windows 10.

    I'll run the benchmarks using Autohotkey and save copies to make sure that the runs are identical. I'm very curious about the amount that the drivers themselves have improved, and the amount changed by the OS.
     
  20. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB

Share This Page