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:
    1,356
    Likes Received:
    14
    GPU:
    Galax 980 ti HOF
    OIh my word check out this. A 4790k vs 5930k even on gtx 980 you see huge improvements in BF4 at least. other games I saw even 10fps increase

    https://youtu.be/CppDFcn8ul8?t=343

    Ok I never thought about this but it could be the recording software that's giving the 4790k a bit slower performance cause they for sure don't have a cam recording the thing. To my understand when recording no way you get 0% performance loss so I bet it's still using some cpu power and the 5930k should help out with that.
     
    Last edited: May 26, 2015
  2. Atlas

    Atlas Maha Guru

    Messages:
    1,370
    Likes Received:
    0
    GPU:
    290 x2, Saphire reference
    The 4790K is faster in TW3 and TR. So it's a throw of the dice.
     
  3. DiceAir

    DiceAir Maha Guru

    Messages:
    1,356
    Likes Received:
    14
    GPU:
    Galax 980 ti HOF
    Yes I saw that. so the 5930k is not really faster even at 4.7GHz. I guess it all depends how the developers implement it. I suspect the make games more optimize for i5/i7 mainstream platform cause that's what most people are running
     
  4. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    7,645
    Likes Received:
    578
    GPU:
    Inno3D RTX 3090
    Most likely it is the recording software. It can also be the game up to a point.

    They mostly optimize for 8 threads, 4 "heavy" and 4 "light" in tandem with the console architecture, making an AMD FX-8xxx or a socket 115x i7 the ideal kind of CPU I guess.
     

  5. Atlas

    Atlas Maha Guru

    Messages:
    1,370
    Likes Received:
    0
    GPU:
    290 x2, Saphire reference
    If the tested games have a cpu bottleneck, which I don't believe, and thus run better on a 5xxx cpu they could rerun the benchmarks on a 10 x 9 resolution. However, in real world results this would amount to nothing since we don't game at that resolution anymore.

    There are some big differences in there anyhow. So it could be the game in question like PRministerGR said.
     
  6. davdes44

    davdes44 New Member

    Messages:
    6
    Likes Received:
    0
    GPU:
    290
    The two comparisons look indiscernible to me
     
  7. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,575
    Likes Received:
    233
    GPU:
    RX 580 8GB
    API overhead test:

    [​IMG]
    [​IMG]
    [​IMG]
     
  8. Vbs

    Vbs Master Guru

    Messages:
    291
    Likes Received:
    0
    GPU:
    Asus Strix 970, 1506/7806
    Hey guys, I've made an updated 15.200.1023.3 driver with the Witcher 3 profile from 15.5 beta (only changed atiapfxx.blb).

    Could any of you test if the profile is working correctly please? (I had to send my cards for a warranty replacement.)

    AMD 15.200.1023.3 (x64-v5-beta).7z
     
  9. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,575
    Likes Received:
    233
    GPU:
    RX 580 8GB
    You're the man! I'll try them ASAP. Good luck with the warranty.
     
  10. sammarbella

    sammarbella Ancient Guru

    Messages:
    3,929
    Likes Received:
    178
    GPU:
    290X Lightning CFX (H2O)
    Thanks GL with the RMA. :)
     

  11. razorbackdb

    razorbackdb Member

    Messages:
    49
    Likes Received:
    0
    GPU:
    AMD R7 260x 2GB
    Is this the app profile or the crossfire profile? or both?
     
  12. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,575
    Likes Received:
    233
    GPU:
    RX 580 8GB
    Crossfire.
     
  13. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,575
    Likes Received:
    233
    GPU:
    RX 580 8GB
    It's not working for The Witcher 3.
    CCC is bugged too. You can't add custom applications anymore.
     
  14. Romulus_ut3

    Romulus_ut3 Master Guru

    Messages:
    673
    Likes Received:
    143
    GPU:
    AMD RX 570 4GB
    Which one of these Windows 10 drivers are the best for gaming?
     
  15. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,575
    Likes Received:
    233
    GPU:
    RX 580 8GB
    1018 or 1023.3
     

  16. Vbs

    Vbs Master Guru

    Messages:
    291
    Likes Received:
    0
    GPU:
    Asus Strix 970, 1506/7806
    Thanks! I think I'll postpone that change to when I get my cards back then. :)
     
  17. bigosik

    bigosik Banned

    Messages:
    161
    Likes Received:
    0
    GPU:
    3072

    WOW, huge improvement ! 500 Draw-Calls more with an old HD7950 (single-GPU), Great Result !

    However , DX12 is the future
    Project Cars patch is coming up to 40% on XBOX and PC !!

    Link :
    http://www.pcgameshardware.de/Projec...splus-1160230/
     
  18. Romulus_ut3

    Romulus_ut3 Master Guru

    Messages:
    673
    Likes Received:
    143
    GPU:
    AMD RX 570 4GB
    I have done 5 run each for the drivers mentioned below with default settings, the results are below.

    Code:
    [LIST]
    [*][B][U]Driver[/U][/B]		[B][U]DX11 ST[/U][/B] 		        [B][U]DX11 MT[/U][/B]
    [*]1012.2
    [/LIST]
    [LIST=1]
    [*]Run 1 		921 402			859 643
    [*]Run 2			920 086			857 034
    [*]Run 3			916 759			858 517
    [*]Run 4			915 212			761 115
    [*]Run 5			919 995			862 038
    [/LIST]
    
    [LIST]
    [*]1018.1
    [/LIST]			
    [LIST=1]
    [*]Run 1			942 772			868 083
    [*]Run 2			893 090			879 295
    [*]Run 3			942 623			874 244
    [*]Run 4			940 495			872 789
    [*]Run 5			945 783			876 372
    [/LIST]
    
    [LIST]
    [*]1023
    [/LIST]	
    [LIST=1]
    [*]Run 1			944 433			877 950
    [*]Run 2			919 903			779 970
    [*]Run 3			938 837			872 909
    [*]Run 4			938 771			869 015
    [*]Run 5			937 029			873 325
    [/LIST]
    
    [LIST]
    [*]1023.3
    [/LIST]
    			
    [LIST=1]
    [*]Run 1			936 832			873 030
    [*]Run 2			935 305			865 872
    [*]Run 3			916 214			866 280
    [*]Run 4			929 373			870 019
    [*]Run 5			939 467			870 302
    [/LIST]
    
    [LIST]
    [*]1036
    [/LIST]	
    
    [LIST=1]
    [*]Run 1			899 158			834 686
    [*]Run 2			892 791			832 739
    [*]Run 3			897 308			834 308
    [*]Run 4			894 257			832 094
    [*]Run 5			870 516			839 108
    [/LIST]
    
    [LIST]
    [*]R9 3XX-Fiji
    [/LIST]
    
    [LIST=1]
    [*]Run 1			938 519			877 759
    [*]Run 2			938 179			878 030
    [*]Run 3			930 424			876 172
    [*]Run 4			934 936			874 963
    [*]Run 5			934 813			876 307
    [/LIST]
    For an even better view:
    [​IMG]
     
    Last edited: May 31, 2015
  19. DiceAir

    DiceAir Maha Guru

    Messages:
    1,356
    Likes Received:
    14
    GPU:
    Galax 980 ti HOF
    My test 1036 vs r9 300 vs 1023.0

    r9 300 driver

    dx11 st = 1 459 754
    dx11 mt = 1 233 189
    Mantle = 16 592 131

    123.0

    dx11 st = 1 446 349
    dx11 mt = 1 316 197
    Mantle = 17 172 870

    1150mhz core 123.0 driver

    dx11 st = 1 459 037
    dx11 mt =1 379 117
    Mantle = 17 631 147


    1036

    dx11ST = 1 471 797
    dx11mt = 1 344 179
    Mantle = 17 257 119

    1150mhz core 136 driver

    dx11 st = 1 361 674
    dx11 mt = 1 360 037
    Mantle = 17 270 481


    running 136 driver 2560x1440 stock clocks

    dx11 st = 1 429 169
    dx11 mt = 1 315 529
    Mantle = 17 141 602


    136 2560x1440 1150 core

    dx11 st = 1 349 683
    dx11 mt = 1 359 760
    Mantle = 17 089 482

    So as you can see it's strange that overlcocking my gpu get me lower scores on the 136 driver but on 123 it's higher when overclocking. I think this test is going nowhere. sometimes better sometimes lower and thing about he 146 driver where I get almost 80k less than stock clocks is not right. Also mantle is lower I'm thinking to myself "how can that be". So to me this test is pointless for now unless you get 300k or more draw calls on dx11 then i see it showing real results. I will still test games to see if there is some real world benifits between older drivers and 136. Maybe I should overclock memory as well but my card doesn't like that.



    Juust one more thing. Everything was tested using radeon pro to force gpu lock speed to max 3d clock and 4.6GHz overclocking on cpu and using performance mode to force it to 4.6GHz during test so nothing was clocked down during test.
     
    Last edited: May 31, 2015
  20. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    7,645
    Likes Received:
    578
    GPU:
    Inno3D RTX 3090
    The 3d mark test seems consistent for my system, almost up to the decimal point at moments, but there is also a 5% of fluctuation. I believe that even the motherboard model/bios plays a role in it.
     

Share This Page