High DX11 CPU overhead, very low performance.

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

  1. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    Thanks!
    No overclock, undervolted slightly with dynamic voltage. Peaks at 3.9GHz as it should. I use Process Lasso for bitsum too.
     
  2. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    You believe that 1023.7 is a keeper over 1040?
     
  3. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    I will install 1040 and give them a test to see which comes out on top. I'm still using 1040 dll's for games though (habit).
     
  4. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    960/1500 Single GPU:

    1040
    [​IMG]
    1023.7
    [​IMG]
     

  5. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    This seems exactly the same frankly. Did the graphics scores change?
     
  6. Gapster

    Gapster Guest

    Messages:
    75
    Likes Received:
    0
    GPU:
    2xSapphire R9 290x Tri-x
    Hi PrMinisterGR, i remember u did many tests about overhead in games and then that thread was deleted. Can u say me what branch of drivers more good for dual-core(4 threads) CPU. Using 15.200.1040, it nice, but i forgot is 14.9-15.6 branch of drivers was better for dual core or not.(of cause if u remember)

    P.S Big thx for answer.
     
    Last edited: Jul 6, 2015
  7. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    For dual core, the newer branch (1018, 1023.5, 1023.7, 1036, 1040, 15.15) were a tiny bit slower. But they were better with frame latencies. I would recommend the newer branch over the old one, no matter the situation.
     
  8. PieEyedPiper

    PieEyedPiper Master Guru

    Messages:
    628
    Likes Received:
    10
    GPU:
    RX 580
    This is somewhat polar from my experiences. But that might have more to do with my hardware and how it is reacting to overhead changes vs. application improvements. For instance, many people are getting higher Firestrike scores with 1023.7 over I guess 1040? (People really need to start saying what driver they're coming from when posting about improvements) but for my system, whatever application level improvements were made don't justify the loss in overhead in comparison to 1040, i.e. my physics score dropped by 100 points due to the less optimized overhead of the 1023.7 drivers. Again, my system is a bit of an odd duck, so this is just my personal experience.

    1040
    DX11 ST 827 849
    DX11 MT 803 808
    Mantle 8 033 265

    Firestrike
    P7085
    gfx8391
    physx7143
    combined3252

    1023.7
    DX11 ST 676 834
    DX11 MT 643 880
    Mantle 8 053 009

    Firestike
    P6965
    gfx8233
    physx7044
    combined3208

    7950 @ 1050/1500
    PCIE 1.1
    LGA775 @ 4.2GHz
     
    Last edited: Jul 6, 2015
  9. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    There's a massive diference there between 1040 and 1023.7 in the overhead test, so strange.
     
  10. vase

    vase Guest

    Messages:
    1,652
    Likes Received:
    2
    GPU:
    -

    all the results, in every thread on these forums show that 1023.7 is obviously optimized for low-level apis
    because mantle and dx12 ALWAYS perform better than in 1040 for example
    and dx11 mostly perform worse
     

  11. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    Yeah, mantle performs better with 1023.7
    Though the DX11 tests here are near identical.
     
  12. PieEyedPiper

    PieEyedPiper Master Guru

    Messages:
    628
    Likes Received:
    10
    GPU:
    RX 580
    Right?!? I figure my system is more sensitive to api overhead than more modern intel platforms and that right there could explain some of the differences in results between driver sets.

    I'm not exactly clear what you mean. From my seat, 1040 is the most overhead optimized driver we have but maybe that's because I'm only concerned with DX11 performance at the moment. Are you saying that 1023.x is always better with regards to mantle/dx12 overhead vs 1040?

    OK, good, I guess I WAS understanding that post afterall ;)
    I still think that your rather "null" result with regards to overhead is because your system is more modern and efficient than my architecture.
     
    Last edited: Jul 6, 2015
  13. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    That could be it, not sure. Big difference though. 1040 is a god send for your system it seems.

    Yeah. 15.15 (R300 drivers) performs a bit worse than 1023.7
    Can you give 15.5/6 drivers a test?
     
  14. PieEyedPiper

    PieEyedPiper Master Guru

    Messages:
    628
    Likes Received:
    10
    GPU:
    RX 580
    15.5
    DX11 ST 593 268
    DX11 MT 564 799
    Mantle 8 342 549

    Firestrike
    P6836
    gfx8024
    physx7106
    combined3155
    15.15 r300
    DX11 ST 672 774
    DX11 MT 660 852
    Mantle 8 051 063

    srry no firestrike
    1040
    DX11 ST 827 849
    DX11 MT 803 808
    Mantle 8 033 265

    Firestrike
    P7085
    gfx8391
    physx7143
    combined3252
    1023.7
    DX11 ST 676 834
    DX11 MT 643 880
    Mantle 8 053 009

    Firestike
    P6965
    gfx8233
    physx7044
    combined3208
     
    Last edited: Jul 6, 2015
  15. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    You have PCI-E 1.1? Are you sure? At what speed? And why so low? Whatever motherboard you have, it must be at least 2.0. I have a separate audio card, and an Intel NIC, both PCI-E, and my gpu is running at PCI-E 2.0 8x. Even then, when the motherboard I have ordered arrives, I will move heaven and earth to make the card run at 16x again.

    There are some real differences when PCI-E link speeds are low, and I have a feeling that the overhead test is very sensitive to that in your system (if you indeed have a system running at PCI-E 1.1, which I find difficult).
     

  16. PieEyedPiper

    PieEyedPiper Master Guru

    Messages:
    628
    Likes Received:
    10
    GPU:
    RX 580
    PCIE 1.1 16x

    Don't forget this is a socket 775 system, bearlake p35 chipset. x38, while indeed socket 775, was the first intel chipset to support PCIE 2.0.
    [​IMG]
     
    Last edited: Jul 6, 2015
  17. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    EDIT: Disregard it, I'm blind. The differences seem minimal (mostly because of the x16 speed), but I don't know about the latencies and the communication with the CPU.
     
    Last edited: Jul 6, 2015
  18. Romulus_ut3

    Romulus_ut3 Master Guru

    Messages:
    780
    Likes Received:
    252
    GPU:
    NITRO+ RX5700 XT 8G
    On Windows 10, I get this:

    [​IMG]

    Uhmm.. WTF?
     
  19. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    What were you getting in 8.1 with the same driver?
     
  20. Romulus_ut3

    Romulus_ut3 Master Guru

    Messages:
    780
    Likes Received:
    252
    GPU:
    NITRO+ RX5700 XT 8G
    Not with this particular driver, as I couldn't test it on Windows 8.1 since I upgraded to Windows 10. But what I was getting is something like this:

    [​IMG]

    Seems on Windows 10, the overhead test behaves similar to the regular Catalyst drivers on Windows 8.1.
     

Share This Page