AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

Discussion in 'Architecture and Products' started by ToTTenTranz, Sep 20, 2016.

  1. lanek

    Veteran

    Joined:
    Mar 7, 2012
    Messages:
    2,451
    Likes Received:
    303
    Location:
    Switzerland
    Fetch need Primitive shader for working or is it different ( called smart primitive on the slide ) .. ? if DSBR is not enabled ( or only case by case and rare case now ), i dont see how it can change ( and vice versa if primitive shader are not active too )
     
  2. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    7,807
    Likes Received:
    2,072
    Location:
    Well within 3d
    That would depend on whether one believes the energy 01 subtest that appears to have DSBR enabled to some extent also has primitive shaders.
    At least in theory, the two features could be orthogonal. However, they can have synergies or might have additional ways internally to have a direct influence.
    Primitive shaders exist in the vertex process step, and at a minimum include better culling before hitting the fixed-function primitive culling and primitive setup.
    It generally appears as if the binning rasterizer exists to consume primitives set up for rasterization, which is part of the stage primitive shaders would be sending data to, or the DSBR is more fully on other side of that stage.

    The DSBR's capacity for binning is very limited based on the number of primitives and their related state, so whatever methods are taken earlier to keep the amount of unnecessary work getting to the binning stage would help.
    Perhaps there could be internal hooks or future optimizations that could allow programmable stages to better inform the binning/rasterization processes of relevant state changes or better handle context management.
     
    DavidGraham, Grall, lanek and 2 others like this.
  3. Digidi

    Newcomer

    Joined:
    Sep 1, 2015
    Messages:
    134
    Likes Received:
    49
    In Energy-01 vega was as fast as P6000/TitanX(p)
    http://www.pcgameshardware.de/Titan...Frontier-Edition-Nvidia-Pro-Features-1234729/
    http://www.pcgameshardware.de/Radeo...6623/Tests/Benchmark-Preis-Release-1235445/3/
     
  4. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    7,807
    Likes Received:
    2,072
    Location:
    Well within 3d
    Digidi and pharma like this.
  5. Digidi

    Newcomer

    Joined:
    Sep 1, 2015
    Messages:
    134
    Likes Received:
    49
  6. lanek

    Veteran

    Joined:
    Mar 7, 2012
    Messages:
    2,451
    Likes Received:
    303
    Location:
    Switzerland
    One "footnote" on the benchmark for CAD software, they are purely indicative and not mean to be a direct comparaison on hardware but more a base.. scores can differ widely from a system, configuration, OS ..and the scene used are offtly not much a good comparaison depending the work you do with the CAD software.. They are offtly not much reflecting last update of the softwares.... ( rarely using the last addition on the softwares and rarely using the last optimization and are offtly old scene or situations more or less obsolete ). So it is a bit a complicated game when we try use thoses benchmark as pure reference ( i mean for office who work in CAD ).. Hence why we offtly have our own test routing.

    Some Linux distro, who are completely empty of just the minimum needed for make work one specific software , can bring 10 to 25% difference for the time needed for a render scene with CUDA and OpenCL vs windows ( let alone OSX ) ( 25% on an non optimized windows .)
     
    #4366 lanek, Oct 12, 2017
    Last edited: Oct 12, 2017
  7. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    7,807
    Likes Received:
    2,072
    Location:
    Well within 3d
    Are you referring to the Creo 3.0 wireframe test?

    I'm not familiar with what that test does differently compared to some of the others. The Beyond3d suite tests for Vega FE showed that it could be competitive or better than the competition in the non-culled and mixed scenarios. The big difference appeared to be in the fully culled cases. Perhaps the workstation test was able to provide a more favorable mix.
     
  8. DavidGraham

    Veteran

    Joined:
    Dec 22, 2009
    Messages:
    1,926
    Likes Received:
    805
  9. CarstenS

    Veteran Subscriber

    Joined:
    May 31, 2002
    Messages:
    4,607
    Likes Received:
    1,780
    Location:
    Germany
    Sorry for the confusion. :(
     
  10. Pressure

    Veteran Regular

    Joined:
    Mar 30, 2004
    Messages:
    1,188
    Likes Received:
    155
    Still waiting on that magic driver, eh?
     
  11. Anarchist4000

    Veteran Regular

    Joined:
    May 8, 2004
    Messages:
    1,335
    Likes Received:
    334
    The funny part about "magic" drivers is that anything beyond 5-10%, maybe 15%, would be a fantasy. Greater than 20% seems to be becoming a trend of late. That argument was also founded on Vega competing with 1080ti, which we've already seen without all the features enabled.
     
  12. Infinisearch

    Veteran Regular

    Joined:
    Jul 22, 2004
    Messages:
    730
    Likes Received:
    135
    Location:
    USA
    What is it competitive with the 1080ti in?
     
  13. Bondrewd

    Regular Newcomer

    Joined:
    Sep 16, 2017
    Messages:
    293
    Likes Received:
    108
    Is there anything magical about enabling marketed uArch features?
     
  14. ninelven

    Veteran

    Joined:
    Dec 27, 2002
    Messages:
    1,680
    Likes Received:
    102
    That depends on the name of the company.
     
    Pete, Putas and pharma like this.
  15. Bondrewd

    Regular Newcomer

    Joined:
    Sep 16, 2017
    Messages:
    293
    Likes Received:
    108
    How the name of the company is in any way relevant to enabling what was advertised (to some degree, that is).
     
  16. Rootax

    Regular Newcomer

    Joined:
    Jan 2, 2006
    Messages:
    741
    Likes Received:
    299
    Location:
    France
    Even without magic drivers, instruction prefetching seems working good on Vega, with good results in Dirt 4 and Forza 7.
     
  17. Malo

    Malo YakTribe.games
    Legend Veteran Subscriber

    Joined:
    Feb 9, 2002
    Messages:
    5,895
    Likes Received:
    1,901
    Location:
    Pennsylvania
    Are there options in those games to turn off instruction prefetching and compare? How is it you know the performance is a result of that?
     
    Grall and pharma like this.
  18. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    7,807
    Likes Received:
    2,072
    Location:
    Well within 3d
    Would Vega's instruction prefetch be different than the instruction prefetch mentioned for Polaris?
     
    pharma likes this.
  19. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,469
    Likes Received:
    1,800
    Location:
    La-la land
    Why would instruction prefetching favor two driving games in specific? Where are the gains for the hundreds of other modern PC games people play? :p
     
    sir doris likes this.
  20. el etro

    Newcomer

    Joined:
    Mar 9, 2014
    Messages:
    95
    Likes Received:
    12
    There will not exist a magical driver. Vega is hardware limited, i say it without doubt. True GCN2 is far from us, still.
     

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...