3dmark TimeSpy scores

Discussion in 'Videocards - AMD Radeon Drivers Section' started by spectatorx, Jul 15, 2016.

  1. Dygaza

    Dygaza Guest

    Messages:
    536
    Likes Received:
    0
    GPU:
    Vega 64 Liquid
    Yeah you shouldn't get invalid score if you disable async. Clearly a mistake from their part. Just like you should be able to get score from different resolutions.
     
  2. Dekaohtoura

    Dekaohtoura Master Guru

    Messages:
    382
    Likes Received:
    56
    GPU:
    XFX 6700XT SWFT
     
  3. SimBy

    SimBy Guest

    Messages:
    189
    Likes Received:
    0
    GPU:
    R9 290
    That would be a very quick&dirty fix, yes. I mean I would still have an issue with that since it would kinda make TimeSpy 'AC' feature pointless but at least there wouldn't be such an obvious discrepancy between what happens if you disable something on a driver level and benchmark level.
     
    Last edited: Jul 17, 2016
  4. pharma

    pharma Ancient Guru

    Messages:
    2,485
    Likes Received:
    1,181
    GPU:
    Asus Strix GTX 1080
    It is very simple as was explained to you in another forum:
    For the life we can't undertand why you don't understand this? Hopefully once FM updates their Technical guide you will have a better understanding...
     
    Last edited: Jul 17, 2016

  5. SimBy

    SimBy Guest

    Messages:
    189
    Likes Received:
    0
    GPU:
    R9 290
    A better fix would be to create separate paths for both Nvidia and AMD and optimize them as much as possible. That way Nvidia would not have to resort to driver hacks to prevent performance loss, you would likely see a much larger gains on AMD hardware and one would expect Pascal to benefit as well.

    And here is why this wont happen:

    FM_Jarnis [developer] 4 hours ago
    "Note, as a benchmark, 3DMark specifically does not have any vendor-specific code path optimizations because those would turn it an optimization contest, rather than hardware benchmark. All hardware runs the same code path. Yes, even on Maxwell. The driver determines how it handles the queues."

    It is kinda funny when you think about. A benchmark that is pretty much all about 'optimization contest' is now somehow above that.

    Not to mention every game has separate code paths. So how exactly is TimeSpy representative of games then.
     
    Last edited: Jul 17, 2016
  6. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    The logic leaps are quite amazing. The result should be invalid if you disable Async, since the game simulates games that will request it. How the GPU benefits from it is a matter of the GPU implementation. Maxwell doesn't, Pascal and GCN do.

    Saying something "simply" doesn't mean it's correct or that it makes any logical sense. See George W. Bush speeches.
     
  7. Dygaza

    Dygaza Guest

    Messages:
    536
    Likes Received:
    0
    GPU:
    Vega 64 Liquid
    In this case, it would actually be a driver hack from nvidia's part to run async code on maxwell that doesn't have native support for it. Seriously, there is no hack.
     
  8. SimBy

    SimBy Guest

    Messages:
    189
    Likes Received:
    0
    GPU:
    R9 290
    Let's just agree to disagree and leave Bush out of it.
     
  9. Redemption80

    Redemption80 Guest

    Messages:
    18,491
    Likes Received:
    267
    GPU:
    GALAX 970/ASUS 970
    SimBy, perhaps message AMD to find out their thoughts on the matter.

    They signed off on everything and the current implementation benefits their hardware more than anyone else.
    Any changes would likely mean lower performance on AMD hardware, why as an AMD card owner would you want that?

    AMD don't need or want people fighting their battles, you just embarrass them and likely have them internally contacting Futuremark to apologise for the grief they are getting.
     
  10. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,125
    Likes Received:
    969
    GPU:
    Inno3D RTX 3090
    I'm leaving Bush out of it, you're right. But it's not simply "agree to disagree" when you compare different things.

    "Agree to disagree" is usually a nice cop out when you can't support what you want to say. Anyway...
     

  11. leszy

    leszy Master Guru

    Messages:
    348
    Likes Received:
    39
    GPU:
    GB 7900XT Gaming OC
    GDC16_gthomas_adunn_Practical_DX12.pdf

    [​IMG]
     
  12. DerSchniffles

    DerSchniffles Ancient Guru

    Messages:
    1,665
    Likes Received:
    148
    GPU:
    MSI 3080Ti
    I got 5148. Not bad I guess?
     
  13. Flano

    Flano Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    2x AMD R9 FURY X 4GB HBM
    8800 - 8900 varies on each run .. painstaklingly close to 9000+ for the acheivement ... bah!
     
  14. ultraex2003

    ultraex2003 Master Guru

    Messages:
    271
    Likes Received:
    93
    GPU:
    Amd 2x290 Cfx
    5820k @ 4.0
    2x290 @ 1150/1550

    [​IMG]
     
  15. elmonen

    elmonen Member

    Messages:
    36
    Likes Received:
    2
    GPU:
    Asus RX6600 Dual
    Cpu @ 4,5Ghz
    Fury @ 1040 / 500

    Graphics score: 4768
    Cpu score: 4646
    4749
     

  16. moaka

    moaka Guest

    Messages:
    291
    Likes Received:
    9
    GPU:
    RTX 2070 Super

    Verry nice 5224 gpu score with only 1105 OC :)
     
  17. C-Power

    C-Power Member Guru

    Messages:
    128
    Likes Received:
    34
    GPU:
    Inno3d 4090 White
    Interesting, it seems the CPU test is a lot less hyperthread optimized as with the Firestrike one, I actually gained CPU score by disableing HT... where on FS cpu test I'd gain like 30/40% fps with HT enabled :3eyes:

    HT Off:
    http://www.3dmark.com/3dm/13420787?

    Cpu score 7293

    HT On:
    http://www.3dmark.com/3dm/13245759

    Cpu score 6708

    Maybe interesting for others to test out - I reckon it's probably optimised for less threads then FS - but I'm no expert :p
     
    Last edited: Jul 19, 2016
  18. Dekaohtoura

    Dekaohtoura Master Guru

    Messages:
    382
    Likes Received:
    56
    GPU:
    XFX 6700XT SWFT
    Anyone care to clarify @what resolution each run/result is?

    If we are all running the free version of TimeSpy, then async is always on, but the resolution of our screen is the resolution of the test (unless I'm completely wrong) so there is no way to have a consistent result db.
     
  19. OnnA

    OnnA Ancient Guru

    Messages:
    17,852
    Likes Received:
    6,742
    GPU:
    TiTan RTX Ampere UV
    Some new OCed RX 480 T.Spy bench:

    [​IMG]
     
  20. miiksu

    miiksu Guest

    Messages:
    136
    Likes Received:
    0
    GPU:
    XFX R9 Nano
    I got it slightly better. GPU score 5358 at @1153/572 :banana:
     

Share This Page