Review: Ashes of Singularity: DX12 Benchmark II with Explicit Multi-GPU mode

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Feb 24, 2016.

  1. Denial

    Denial Ancient Guru

    Messages:
    13,527
    Likes Received:
    3,073
    GPU:
    EVGA RTX 3080
    Maybe he looked at the code, like you looked at the architecture.

    Why is this funny? Did anyone even bring up the FCAT results besides you? I think one other guy did and all he asked was that Hilbert show to 980 cards. Everyone else, including me, probably just read the article and assumed Hilbert was right and that it wasn't a big deal. No one even mentioned the ET article until you did. So why would anyone defend something that was never an issue to begin with?

    I don't even have an issue with people hating on Nvidia. As I said multiple times, even before you strawmanned my arguments, Nvidia definitely has an Async problem they need to fix. If there is a driver coming, it should have been out already. It's not really acceptable.

    The thing I do have an issue with is people posting flamebait crap like the first two posts you had in this thread. The "Fanboys" one and the "LOL IMAGE LINK" one. They serve no purpose in the thread other then to provoke people. That's what flamebaiting means. I don't care if you're right, it's how you posted it. The other problem is the crap I quoted on the top of this post. You're asking him to prove something while at the same time, claiming that you know the driver won't add performance because you apparently looked at the architecture and were just somehow able to glean that information from it. Even more amusing is that you say that, right after Ext3h, who is a better source then you, says it's basically all speculation at this point. That's not even to mention your 2019 argument, because apparently you've also looked at the Pascal architecture before anyone else on the entire internet.

    Honestly, like I said, it's definitely possible that you're correct about everything. But at this point it's just an assumption. And using words like "impossible" is misleading nonsense.
     
    Last edited: Mar 3, 2016
  2. mR Yellow

    mR Yellow Ancient Guru

    Messages:
    1,935
    Likes Received:
    0
    GPU:
    Sapphire R9 Fury
    I see a new update is out.
     
  3. GeniusPr0

    GeniusPr0 Maha Guru

    Messages:
    1,314
    Likes Received:
    31
    GPU:
    RX 6800 XT Black Ed
    Read: https://developer.nvidia.com/dx12-dos-and-donts

    Realize that they have 1 work distributor, and a capacity for 31 compute commands.

    Now look at the game and realize with the update that it's a more saturated use of Async Compute. You can see this because the FuryX gains 20% performance at 4K. The load is light, at low res, which might explain its performance regression, for now I'm not sure. Great nVIDIA perfomance at 4K is just never going to happen. 2 FuryX in CFX net 60FPS avg in 4K.


    Relevant disclaimer:

    - Oxide

    Oh, look what I found:

    So there's AMD, with their 2 ACES or 8 ACES, that can truly provide breathing room, and won't run into these problems.

    My overall opinion is that async compute is a disaster in terms of development. If I was a developer, I would probably opt to just not use it, not until I could find a way that didn't seem gimmicky. It's not worth the trouble.


    Nope, wasn't me. I did not bring it up.

    Ext3h has actually evolved his opinion over the past half year or more ;).

    Never trust a marketing team to say the full truth. By not supporting async compute, Robert means supporting it like how AMD is, without any driver intervention on the DX12 side, while doing both graphics and compute.
     
    Last edited: Mar 4, 2016
  4. chispy

    chispy Ancient Guru

    Messages:
    9,090
    Likes Received:
    1,336
    GPU:
    RX 6900xt / RTX3090
    Yep i saw that too.
     

  5. -Tj-

    -Tj- Ancient Guru

    Messages:
    17,236
    Likes Received:
    1,955
    GPU:
    Zotac GTX980Ti OC
    btw, async compute has already been in COD Ghosts @ DX11 for a while, so its nothing completely new.

    seta r_asyncCompute "1"

    for
    seta r_apex_turbulence "1"
    seta r_reactiveTurbulenceEnable "1"
    seta r_fur_shader "2"
    seta r_useComputeSkinning "1"


    Just saw it now, when I was editing few settings in mp.cfg
     
  6. xIcarus

    xIcarus Master Guru

    Messages:
    954
    Likes Received:
    96
    GPU:
    1080 Ti AORUS
    Why? Because it addresses a different use case?

    How the hell does that make sense? How many games do you expect will use such huge amounts of async like AotS?

    That's not what I implied and you know it. It's one benchmark - one game.

    Maybe because Oxide has been working with AMD? I never said AMD wrote the code themselves - in fact in case you haven't noticed AMD seems to be outing a lot of stuff through the magical open-source portal and expecting others to implement their designs. Their 3D solution is where that went wrong; it's been largely abandoned. However async is an example where it went right.

    Or do you believe Oxide would implement AotS the way Nvidia wants it? Get real.


    *skipped a few sections of you quoting me and replying something completely random which has nothing to do with what I said*


    Oh you work for Nvidia now? You know what they know? I'll be damned!


    False, in the future the architectures will be more similar in order to cater to the trends that DX12 and Vulkan are starting or will start in the future. It is in fact very possible to standardize async implementations in such a way that the hardware respects the specification after the code. This way, the code will constantly be improved in a vendor-agnostic manner.
    History has proven that this is always the case.

    Not ridiculous?

    Should I draw the conclusion that Crapworks' tessellation factors are not overkill because Maxwell can handle them?

    You mean 'since we know it was already enabled, despite the fact that it runs like crap'.

    --

    You think you know something when in fact you don't. Your arguments are based on way too many assumptions and you're either blind to the obvious or deliberately ignoring it to your advantage.

    Ain't even going to care to address your silly personal attacks against me or Denial. In fact, don't expect another reply from me. I'm just beating a dead horse talking to you.
     
    Last edited: Mar 5, 2016
  7. Alessio1989

    Alessio1989 Ancient Guru

    Messages:
    2,114
    Likes Received:
    634
    GPU:
    .
    DirectX 11 does not allow explicit control of the command queues, since commands are executed on a single immediate context.
     
  8. Dygaza

    Dygaza Master Guru

    Messages:
    536
    Likes Received:
    0
    GPU:
    Vega 64 Liquid
    That is most likely for physx. Nvidia can do async compute through cuda.

    "Call of Duty: Ghosts provides a more immersive gaming experience through the use of NVIDIA GameWorks technologies: interactive smoke (NVIDIA Turbulence), dynamic fur simulation on Riley and the wolves (NVIDIA HairWorks) as well as TXAA."
     
  9. Lane

    Lane Ancient Guru

    Messages:
    6,361
    Likes Received:
    3
    GPU:
    2x HD7970 - EK Waterblock
    Lol, yes dont read too much on the term used in COD ini.. this is not the same thing. Its on the Kernel side and only concern compute queue. (At the difference of DX12, Vulkan ( Mantle ) features where the point is to launch asynchronously graphic and compute queue in a non sequential series of call/command automatically ( Without having the driver who choose when and where to send it )

    Its one reason why APEX is flawed on CPU side.. it allways launch the different kernels call / command sequentially. Allways waiting that the first have finish before launch the second one. Creating an extreme CPU thread Bottleneck. Its not that it is multi-threaded or not, the question is just that even till multi-threaded it is still sequential.
     
    Last edited: Mar 5, 2016
  10. -Tj-

    -Tj- Ancient Guru

    Messages:
    17,236
    Likes Received:
    1,955
    GPU:
    Zotac GTX980Ti OC
    Well yeah and for compute textures.. thats why i added those which would affect it.

    And its not any different in dx12 still using cuda. And i really doubt pascal will change this, its how they build it.
     

  11. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    Can't believe this discussion hasn't moved an inch since I last posted. Same points being made, again, and again, and again.

    There are two things that are being overlooked by both 'sides' of the argument :

    1. There is so indication that 'async' will be beneficial to performance on maxwell/kepler

    2. Async compute is a misnomer ; what everyone is screaming about is actually concurrent execution of graphical and compute workloads

    The Dx12 spec does not require concurrency, only asynchronous execution which has been available for a while, but not standardized as part of dx

    Denial is insisting that the NV implementation is broken,not because it performs badly, but because it's detrimental to performance.

    Games that don't use 'async' extensively won't saturate gcn

    Essentially what I'm saying is that is concurrent execution only helps when you can't saturate the gpu otherwise, which isn't the case on maxwell

    You can of course continue to have this argument, built on the wrong premise, but wouldn't it be best for everyone if we all abandoned our 'opinions' (that shouldn't exist) and look at facts.
     
  12. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    Developers need to invest in maximizing throughput on gcn with concurrency, that same investment can be applied towards optimizing for maxwrll/pascal.; this is just another episode in the soap opera that has become of consumer electronics
     
  13. -Tj-

    -Tj- Ancient Guru

    Messages:
    17,236
    Likes Received:
    1,955
    GPU:
    Zotac GTX980Ti OC
    Do's and don'ts in DX12 on nvidia HW..

    It has a lot to do how a game developer now uses its features, dx11 driver did a lot if not all the work, now its all up to developer.


    https://developer.nvidia.com/dx12-dos-and-donts


    it has async and all that parallelism like its been said,
    [​IMG]

    but in this particular AoS benchmark it appears as dev has been ignoring certain dx12 api boundaries or using them in inefficient way across the board..
    In other words sloppy DX12 usage, just like other dx12 amateur benchmarks out atm.


    EDIT: bellow NV gpu + intel gpu @ DX12 U4E elemental, ok its just 4fps extra.. but still something

    So this combo is possible just fine after all :nerd:
    https://blogs.msdn.microsoft.com/di...p-dormant-silicon-and-making-it-work-for-you/


    [​IMG]
     
    Last edited: Mar 10, 2016
  14. chispy

    chispy Ancient Guru

    Messages:
    9,090
    Likes Received:
    1,336
    GPU:
    RX 6900xt / RTX3090
    Would be nice if Hilbert could do an update with the newest drivers from both camps Nvidia and AMD , i would like to see if any improvements have been made on the latest drivers.
     
  15. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    1440p HIGH settings, 980Ti @ 1500/3800 ForceWare 365.51
    [​IMG]

    1440p CRAZY settings, 980Ti @ 1500/3800 ForceWare 365.51
    [​IMG]
     
    Last edited: Mar 16, 2016

  16. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    These are my benchmark results at 1440p CRAZY preset, with and without 4xMSAA, on the last BETA of the game. I'm about to test release version right now

    980Ti@ 1490/8000, 5820k@ 4.5GHz, 32GB DDR4
    4x MSAA CRAZY 1440p
    [​IMG]


    NO AA - CRAZY 1440p
    [​IMG]

    Release Version of the game:
    4xMSAA CRAZY 1440p
    [​IMG]
    NO MSAA- CRAZY 1440p
    [​IMG]
    NO MSAA CRAZY DX11 1440p
    [​IMG]
    EXTREME PRESET 1440p
    [​IMG]
    EXTREME PRESET DX11 1440p
    [​IMG]
    HIGH PRESET 1440p
    [​IMG]
     
    Last edited: Apr 2, 2016

Share This Page