AMD Fixes More Ryzen Issues with New Microcode

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Mar 31, 2017.

  1. moeppel

    moeppel Member Guru

    Messages:
    153
    Likes Received:
    23
    GPU:
    1080 Ti
    You have missed it yet, but AdoredTV (and now C't) has done some investigation and testing.

    Turns out that Ryzen delivers unbelievable performance with AMD cards and the bottleck we have been seeing with Nvidia cards (i.e 1080(Ti)) are not in fact a CPU bottleneck but one of Nvidia's DX12 driver implementation(s). :infinity:

    EDIT: http://i.imgur.com/OjSIcgM.jpg

    Effectively reducing the gap between Ryzen and other CPUs we've been seeing in reviews. Of course, Intel remains champ ;)

    Creates some hype for Vega, if nothing else and highlighted that perhaps the fastest single GPU of both vendors should be tested.
     
    Last edited: Apr 1, 2017
  2. Yxskaft

    Yxskaft Maha Guru

    Messages:
    1,464
    Likes Received:
    115
    GPU:
    GTX Titan Sli
    The point is that when testing the GTX 1070 + Ryzen in DX12, it simply looked like a CPU bottleneck for the Ryzen.
    But when testing RX 480 Crossfire + Ryzen, DX12 performance almost doubled compared to the GTX 1070. Had it been a simple CPU bottleneck, the performance for the RX 480 CF would have been similar to the GTX 1070's result.

    I think he should have tested at 720p to minimize the GPU bottleneck though, and with a single RX 480.
     
    Last edited: Apr 1, 2017
  3. Loophole35

    Loophole35 Ancient Guru

    Messages:
    9,793
    Likes Received:
    1,148
    GPU:
    EVGA 1080ti SC
    I believe you are reading that chart wrong in all the games when tested with the FuryX the 1800x is still behind. The FuryX is not as fast as the GP102 but the two cards are effectively even on 1800x.
     
  4. moeppel

    moeppel Member Guru

    Messages:
    153
    Likes Received:
    23
    GPU:
    1080 Ti
    Check the 720p benchmarks and compare the 1800X with Nvidia Titan X (Pascal) to the Fury X results. (the green and red highlighted ones)

    Occasionally, in 720p, the 1800X + Fury X will outdo the 1800X + Titan X.

    That effectively shows that the bottlenecks we've been seeing aren't in fact CPU bottlenecks. Technically the Fury should easily be 50% less powerful than a Titan on average (if not more so) but manages to tip the tide into its favor - and not just marginally.

    Thus the bottleneck presented by 1800X + Titan X Pascal is likely more a bottleneck of Nvidia's API implementation.

    In Ashes the Fury X is roughly 17% faster (min + max) and RoTR almost 30% (max) and 19% avg.

    Even in GTA 5 (DX11) the Fury magically beats the Titan ~16% (max FPS)

    In all scenarios I'd expect the Fury to be equal to the Titan if Ryzen was the CPU bottleneck but never the better performing card.
     
    Last edited: Apr 1, 2017

  5. JamesSneed

    JamesSneed Maha Guru

    Messages:
    1,151
    Likes Received:
    493
    GPU:
    GTX 1070
    Nvidia driver / API bottleneck

    That is pretty interesting that the large differences in some games could be the Nvidia Directx 12 driver / API. Hilbert do you have a pair of 480's to test this out? Curious how BF1 behaves running Directx12 on ryzen vs a 7700K with Nvida vs AMD graphics cards.

    Funny people have been saying all long you should test at lower resolutions which is very reasonable however if Nvidia has a Directx 12 driver bottleneck that is pacing the frame rate the lower resolution would favor the fastest single threaded performance regardless if the game is multi-threaded.
     
  6. Loophole35

    Loophole35 Ancient Guru

    Messages:
    9,793
    Likes Received:
    1,148
    GPU:
    EVGA 1080ti SC
    No the more powerful the card the harder the bottleneck. These charts still show Ryzen as a bottleneck.
     
  7. TieSKey

    TieSKey Member Guru

    Messages:
    187
    Likes Received:
    66
    GPU:
    Gtx870m 3Gb
    Lol. By your "logic", testing a Ti X stock vs a Ti X overclocked should shield lower FPS for the later?????
    A bottleneck is a bottle-neck, it doesn't matter how big the bottle is.
     
  8. Denial

    Denial Ancient Guru

    Messages:
    13,326
    Likes Received:
    2,827
    GPU:
    EVGA RTX 3080
    Idk, I watched Adored's video and I guess I agree that there is definitely something going on with Nvidia's driver, in DX12, on Ryzen specifically. I know TechEpiphany did Division, but they didn't test Nvidia's DX12 performance on Intel.. they just showed it lower than Polaris on Ryzen - which I mean is still bad, but my DX12 performance in Division is worse than DX11 on my 4790K.

    So for right now, this issue might be specifically limited to ROTTR on DX12 on Ryzen. I haven't seen any other video/site test any other DX12 game to the same degree.

    Either way, Ryzen is still pretty new - and we have no idea how long Nvidia has had Ryzen test setups to do bring up on their drivers. Obviously AMD will have a more mature Ryzen driver, as they built the CPU and had engineer samples/setups for ~6 months now. If in a year it's still a problem, then yeah Nvidia is dropping the ball. But if it gets fixed in like ~6months then I don't see the issue. It's really no different than all the games that need to be patched for performance with Ryzen.
     
  9. Loophole35

    Loophole35 Ancient Guru

    Messages:
    9,793
    Likes Received:
    1,148
    GPU:
    EVGA 1080ti SC
    Look at the chart you posted. Most of the 720 results match the 1080 results with the Fury X. There is a bottleneck with both. Maybe Nvidia needs to update their driver and optimize it for Ryzen but appears AMD does as well.


    Edit: just looked at the charts again I don't read German so is there a reason he omitted 1920x1080 results with the Fury X card. That throws a red flag up for me.
     
    Last edited: Apr 1, 2017
  10. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    7,687
    Likes Received:
    613
    GPU:
    Inno3D RTX 3090
    What about the fact that GPU drivers are basically just in time compilers that are heavily depended on the CPU itself? I know you said that you don't take AdoredTV's videos seriously (and neither you should, especially the "predictions" ones), but his Tomb Raider video looked very compelling to me. AMD would have tuned their driver to their CPU architecture, NVIDIA is probably still doing it right now.

    If someone could debunk it without "Adored lulz", I would be very happy.
     

  11. Silva

    Silva Maha Guru

    Messages:
    1,468
    Likes Received:
    618
    GPU:
    Asus RX560 4G
    I agree with your point and I'm sure every tech reviewer thought exactly the same.

    That doesn't exclude the fact that everyone knows for months that DX11 and DX12 have different performance with AMD and Nvidia and no one bothered to test and hammered AMD for launching crap CPU's again.

    How's that good journalism?

    It's possible to test software and drivers with a RX480 without having GPU bottlenecks.

    I agree this topic needs further investigation from people who actually have the hardware to test :nerd:
     
  12. Aura89

    Aura89 Ancient Guru

    Messages:
    8,161
    Likes Received:
    1,274
    GPU:
    -
    With how these threads have been going lately, it was 99% likely you were being serious. Might want to end with something that shows you are joking, can't read humor in text without obvious humor in text. :p
     
  13. TieSKey

    TieSKey Member Guru

    Messages:
    187
    Likes Received:
    66
    GPU:
    Gtx870m 3Gb
    It wasn't me who posted the charts and I don't know nor actually care who has a bottleneck. I just felt the urge to point out how flawed that line of reasoning was :p
     
  14. Loophole35

    Loophole35 Ancient Guru

    Messages:
    9,793
    Likes Received:
    1,148
    GPU:
    EVGA 1080ti SC
    Well the charts I was referring to showed multiple Intel CPU's and the 1800x tested in 720, 1080 and 1440 with a TXP and just the 1800x with a Fury X tested in 720 and 1440. Now with that limited about of info people are coming to a conclusion that it's a big conspiracy against AMD. Do you. It find it funny that 1800x/FuryX 1080 scores were omitted and no scores with Intel CPU's and the Fury X were shared?
     
  15. TieSKey

    TieSKey Member Guru

    Messages:
    187
    Likes Received:
    66
    GPU:
    Gtx870m 3Gb
    And what would that data show us?

    It's irrelevant. If u use a CPU at 100% with 100% efficiency, changing the GPU hardware can only make u go down, not up, cuz CPU is already at 100% doh...
    So "IF" the charts are "true", and using a less powerful GPU (with a different driver), actually increases FPS, it must mean that the CPU wasn't used
    at 100% efficiency.
    So, thinking nvidia drivers are/were un-optimized for Ryzen is a sound conclusion.
    Why should we assume that those nvidia drivers were optimized since day 1 (when most reviews were done) when even AMD is still ironing the bios and only 2 games released an optimization patch so far?

    The point of the guy who posted the charts was exactly that.

    And I too would like to see some updated reviews from HH, with not only new bios but some AMD GPUs.
     

Share This Page