High DX11 CPU overhead, very low performance.

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

  1. Spartan

    Spartan Guest

    Messages:
    676
    Likes Received:
    2
    GPU:
    R9 290 PCS+
    Yep, I tested mantle with Star Swarm. However, it's almost pointless, due to only a very few games support mantle, and if this will be the case with dx12 as well, in 2016, then it doesn't really matter how amd cpus "should" perform.
     
  2. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    X1 will have Windows 10 and DX12. It's not going to be anything like Mantle :D
     
  3. Dygaza

    Dygaza Guest

    Messages:
    536
    Likes Received:
    0
    GPU:
    Vega 64 Liquid
    It's quite interesting to see how driver behaves for each of us. Could you actually run that test again with 14.12 dll, and confirm your 630k score is still valid?

    For me the benefits have been 920k to 1.3M (14.12 to 1055). So for me the benefits have been exactly 41%.

    Note, we have different windows versions, so there can be different optimisations between WDDM 1.1 and 1.3, which would explain different benefits between different Windows versions.
     
  4. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    Unless there are games that demand much more than the current consoles can give (and the only one I can think of right now is Star Citizen), all AAA titles will be cross platform, optimized for a lot of little x86 cores and GCN. Combine that with a low level API, and I'm sure that a 270x and an fx-8120 can give you better than PS4 performance for everything. How is that bad? Consider the costs of said hardware, and what it enables.

    I just tried that. 3DMark simply gives me a black screen when I try to run the overhead test with the 14.12 DLL. I have always been around 600-650k. Unfortunately I don't have any screens of that, since the 3DMark overhead test came out after the whole mess with the DX11 overhead thread. I was testing only games back then.
     

  5. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    I will test 14.12 again and report back results vs latest 1055 dll's
     
  6. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    old14.12 drivers i get 1.1M and with new drivers 1.4M. For the lols with 12.6 drivers i only get 900K. That's so low and even with crossfire
     
    Last edited: Jul 21, 2015
  7. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    You basically have two cards like mine, and a bit faster CPU. I clearly remember scoring in the 600ks with the 14.12, and it seems ok.
     
  8. DiceAir

    DiceAir Maha Guru

    Messages:
    1,369
    Likes Received:
    15
    GPU:
    Galax 980 ti HOF
    600k sounds about right to me. This shows that driver still pays a big part in performance
     
  9. Spartan

    Spartan Guest

    Messages:
    676
    Likes Received:
    2
    GPU:
    R9 290 PCS+
    When was the last time you played a game with an fx cpu? Ofc, I can play 2012-2013 titles without major problems, but when I try to play a 2015 title, well, that's an entirely different story... Like a rollercoaster, even with downgraded graphics. Current solutions: cap the game at 30 fps or lower the graphics below medium, or simply don't play the game, but seriously, who pays 3 times of a ps4 (my situation) to play games below ps4 quality, or in worst case scenario with xbox360 quality. Well, the answer is me...
     
  10. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    Windows 10 single GPU 960/1500

    15.7 (15.7 main driver)
    [​IMG]
    1055 (15.7 main driver)
    [​IMG]
    1046.2 (15.7 main driver)
    [​IMG]
     

  11. Yecnot

    Yecnot Guest

    Messages:
    857
    Likes Received:
    0
    GPU:
    RTX 3080Ti
    Sir, you need to calm down.

    Look at every Techspot cpu benchmark since forever and tell me if them *****s be cappin' **** at 30fps with an fx 8350. Fx might be slower than the i5 route, but its not getting any slower, and i5s aren't getting any faster tbh.



    Stopping circlejerks in their tracks since sentience :blunt:
     
  12. Gapster

    Gapster Guest

    Messages:
    75
    Likes Received:
    0
    GPU:
    2xSapphire R9 290x Tri-x
    New results on my laptop(i3-2330m@2.2ghz and 6650m@6730m):
    [​IMG]
     
  13. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    Looks good.
     
  14. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    My brother has a little system with an FX-6300, 8GB RAM and a 750Ti. He plays at 1680*1050, and he plays everything at either medium/console settings at a very easy 40+fps (sometimes even double the console fps, at around 60), or he can crank up visuals (not everything) and play at a locked 30. He uses the same machine for work, movie watching, CounterStrike and LoL. He has literally zero problems with it. You have your own experiences obviously, but out of curiosity, which was the title that you were CPU bound before you were GPU bound?

    So, we have reached 1.2m draw calls in DX11 ST with a stable driver. Nvidia is still the leader with 1.9/2m more or less. They still need to improve a lot.

    What I get out of this (unless I'm blind) is that the differences are minimal for you, right?
     
  15. Spartan

    Spartan Guest

    Messages:
    676
    Likes Received:
    2
    GPU:
    R9 290 PCS+
    In average fps it's not that bad, however the main problem with it, this cpu bottlenecks so bad and so often, it's barely usable in games.

    CS (if you mean GO) has an 11 years old engine and LoL isn't a demanding game either.

    I don't know... Maybe in every game? And I'm not talking about Project Cars or Arkham Knight or something like those (my god, I cannot even imagine how bad would they run on my rig), I'm talking about games like Gta V or Witcher 3 (yep, even gta is still using 1 main thread with some supportive secondary threads). Actually, the only 2015 title where I was able to maintain stable 60 fps was MKX (not in cutscenes/x-rays/etc. ofc), after a few patch, with downgraded graphics, 1080p, on a single monitor. But YAY, at least my cpu is capable to run a 2D-3D game with up to 2 characters on screen! :banana:
     

  16. MatrixNetrunner

    MatrixNetrunner Guest

    Messages:
    125
    Likes Received:
    0
    GPU:
    Powercolor PCS+ R9 270X
    Did you try to see what exactly is holding you back? Is it really CPU utilization, some background processes or a slow HDD? On paper you should have a pretty decent rig.

    Did you check latencies? Perhaps you have a problematic driver (SATA, LAN, sound) that blocks CPU access periodically?

    Do you turn off browsers while gaming?
     
  17. Spartan

    Spartan Guest

    Messages:
    676
    Likes Received:
    2
    GPU:
    R9 290 PCS+
    My CPU doesn't exceed 50% in games usually, sometimes, in some loading screens and so, it can reach 100%, but not in actual game. Ofc, there is some background processes, which increase my cpu usage above 50%. I actually tried to set cpu affinity with Process Lasso, I tested different patterns, like 1-3-5-7, 0-2-4-6, 0-3, etc. for both the running game, and for background processes, without any luck. I have a Kingston 120GB SSD, a 1 TB WD Blue Sata 3.0, and a 500 GB Seagate Barracuda HDD Sata 2.0. I tried to run games from all of them, but I have the exact same performance in games (even with seagate sata 2.0), the only noticeable difference is that, the loading times are somewhat faster on the ssd. Also, I don't get any kind of micro stuttering in games, with only one exception, which was The Vanishing of Ethan Carter, but that was a f*cked up game.

    How can I check this?

    Yes.

    However, I don't thing my rig has some kind of problem (except that, I have a pos cpu), because I compared my benchmark results (for example Valley) on the internet, with other oc'd fx8350/r9 290 combos, and got similar results, and I also checked youtube videos with similar rigs, and they have the same "automatic bullet time" feature, or simply play games with 30-35 fps...
     
  18. MatrixNetrunner

    MatrixNetrunner Guest

    Messages:
    125
    Likes Received:
    0
    GPU:
    Powercolor PCS+ R9 270X
    http://forums.anandtech.com/showthread.php?t=2250925

    Here are some links about the latency problems and the tools you can use to check your computer.

    PS. What are big AAA games from 2015, except GTA V and The Witcher 3? And they are still being patched (we can disregard Batman for the moment).
    Can the problem be stated like this: All AAA cross-platform games from the current generation of consoles (PS4 and XBOne)? Or you have problems with specific games that are problematic for most people?
     
    Last edited: Jul 28, 2015
  19. Spartan

    Spartan Guest

    Messages:
    676
    Likes Received:
    2
    GPU:
    R9 290 PCS+
    Well, I disabled my audio device, and get somewhat better results with latencymon, however I didn't notice any difference in gta 5, but probably will test this thing further, thanks!

    Duno... Maybe... I guess...
     
  20. OneB1t

    OneB1t Guest

    Messages:
    263
    Likes Received:
    0
    GPU:
    R9 290X@R9 390X 1050/1325
    its single thread performance bottleneck :) i hope that DX12 games will come ASAP
     

Share This Page