RX Vega Owners Thread, Tests, Mods, BIOS & Tweaks !

Discussion in 'Videocards - AMD Radeon' started by OnnA, Aug 15, 2017.

Thread Status:
Not open for further replies.
  1. OnnA

    OnnA Ancient Guru

    Messages:
    9,733
    Likes Received:
    2,046
    GPU:
    Vega 64 XTX LiQuiD
    Max alowed (I mean resonable) Hotspot is ~80-85deg (so if You have 105deg then we got a problem)
    Try to add some Fan to the Mix (to chill this bad boy a little)
    Start with simple solutions.
    We don't like Temps beyond 75deg !

    Note:
    Not any temp should be above 80deg IMhO.
    I have Max HotSpot at ~70deg (GPU & HBM2 ~60deg.) when Playing SotTR in DX12.
     
    Last edited: Dec 14, 2018
  2. slicer

    slicer Member Guru

    Messages:
    109
    Likes Received:
    34
    GPU:
    Gigabyte VEGA 64
    I have heard that HotSpot temp is some arbitrary number that doesn't actually do anything if it is over 100C.
     
  3. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    15,728
    Likes Received:
    1,718
    GPU:
    Sapphire 5700XT P.
    There's a lot of guesses and estimates but AMD themselves are silent about it. I've never found anything conclusive so far but I'd ignore it although 105c is where it starts throttling if this value is hit whatever it's calculated from.
    Gigabyte is the OC model which is a bit average for the third party models isn't it? Not as problematic as the early Asus cards though I believe but I think the later revisions resolved the thermal pad fittings to be better. :)

    Even Sapphire hits problems occasionally though, some users reported higher than expected temps sometimes almost to the limit of the GPU being able to handle it and then it turned out to be a warp or bend in the heatsink causing it so possibly a bad batch or a few misses during the testing phase before it got packed and shipped to store.

    EDIT: Hmm I might be thinking of the MSI Air Boost and not the Gigabyte model. :)
     
  4. nevcairiel

    nevcairiel Master Guru

    Messages:
    614
    Likes Received:
    199
    GPU:
    MSI 1080 Gaming X
    Its not going to happen though. If this was possible with current shader hardware, someone would've done it already. 3D developers have been waiting and hoping for this for years, if not decades.
     

  5. G13Homi

    G13Homi Active Member

    Messages:
    73
    Likes Received:
    23
    GPU:
    Vega 64 2x xfx
    My bios has a limit of 85* anything above that and i get the overheat crash screen ( blue screen with a rainbow horizontal line) thats a reference bios though.
    If that was my card, i would either request a refund or return for overheating, or take it apart and reseat the thermal pads and replave the paste. Ive had to do this on a couple cards in the past, No big deal.
     
  6. Exodite

    Exodite Ancient Guru

    Messages:
    1,913
    Likes Received:
    171
    GPU:
    Sapphire Vega 56
    Early impressions based on preliminary testing (~4 hours of various 3DMark runs) with 18.12.2 and using WattMan to set p-state clocks and voltages instead of softmod indicates that, for me;
    • voltages drop lower than set values using WattMan in 18.12.2 than using softmod on 18.11.2, leading to instability without further tweaking.
    • memory seems more stable, I've now been able to move back to 950MHz from 875 (@950mV) but I need to test this thoroughly given my previous instability.
    • the resting voltage of the card is reporting lower values on 18.12.2, ~750mV as opposed to ~775-800mV on 18.11.2 and ~850-900mV with softmod. I've yet to check if this impacts idle power consumption.
    • exiting 3D mode, like coming out of a fullscreen benchmark/stress test or game, gives a momentary black screen which leads me to speculate that some low-level functions are being loaded/unloaded based on 3D load. Obviously it could be something else entirely. :)
    In conclusion I've been forced to increase some voltages, and decrease p7 clocks, compared to my softmod setup which results in slightly lower performance (not measured fully yet). Not having to rely on 3rd party solutions to manage p-states is, for me, worth the tradeoff.

    Unfortunately I never bothered to track power usage and temperatures with Chill before so I have no basis for testing the improvements.

    All voltage readings based on GPU-Z logs so these are far from conclusive, and reporting could differ based upon the low-level the changes in the 18.12.2 driver set perhaps?
     
    Jackalito and Dekaohtoura like this.
  7. OnnA

    OnnA Ancient Guru

    Messages:
    9,733
    Likes Received:
    2,046
    GPU:
    Vega 64 XTX LiQuiD
  8. OnnA

    OnnA Ancient Guru

    Messages:
    9,733
    Likes Received:
    2,046
    GPU:
    Vega 64 XTX LiQuiD
    My New Best (with 1.150mV 1790MHz/1200HBM2 Tess x8)
    Real MHz is a little lower (~1750-1760)

    -> https://www.3dmark.com/3dm/31281563?

    [​IMG]

    @Ernwild108 from overclock.net
    GPU score at almost 29k (28.755)

    [​IMG]
     
    Last edited: Dec 15, 2018
  9. Dekaohtoura

    Dekaohtoura Master Guru

    Messages:
    282
    Likes Received:
    30
    GPU:
    Sapphire V56 Pulse
    Nice findings.

    Your Sapphire 56 is a Pulse, right?

    Since you're mentioning HBM voltage, is it possible now to set it from within Wattman?
     
  10. Exodite

    Exodite Ancient Guru

    Messages:
    1,913
    Likes Received:
    171
    GPU:
    Sapphire Vega 56
    Indeed, I'm using a Pulse!

    As for HBM voltage then no, AFAIK you can't set the actual HBM voltage with anything short of a BIOS mod. Can't softmod it either.

    I meant what shows as memory voltage in the application, ie. SoC voltage floor or whatever it is. Previous testing has shown a tendency towards additional instability if I lower it and little or no gain by increasing it so I tend to leave it at default (950mV).
     
    Jackalito and Dekaohtoura like this.

  11. Dekaohtoura

    Dekaohtoura Master Guru

    Messages:
    282
    Likes Received:
    30
    GPU:
    Sapphire V56 Pulse
    For some curious reason, it's at 930mV on my card...I'll set it back to the degault value and see where this gets me.

    I wonder if this causes some minor instabilities that I've had.
     
  12. OnnA

    OnnA Ancient Guru

    Messages:
    9,733
    Likes Received:
    2,046
    GPU:
    Vega 64 XTX LiQuiD
    OP Updated

    Registry MOD v.004
    Just merge with registry (Do backup)
    Need to do it always after new driver install.

    Note:
    Make sure to look into it with Notepad++ and adjust some options
    Im using 1440p 70Hz Monitor, so many new resolutions are with 70Hz.
    "DALNonStandardModesBCD1"=

    -> https://mega.nz/#!4I9lUKza!RoVp71cDoAytQJ5ZEcq3bIM2-N6j4EAAaQVtjihlHy0
     
    Last edited: Dec 15, 2018
  13. mtrai

    mtrai Maha Guru

    Messages:
    1,125
    Likes Received:
    332
    GPU:
    PowerColor RD Vega
    @OnnA have you thought about trying out for you Soft PP tables?

    Vega Soft Power Play Editor
    Gtk# for .NET

    Been messing with it today finally...and actually gives a lot more fine control then overdriventool especially with the Adrenlin 2019 drivers.
     
  14. OnnA

    OnnA Ancient Guru

    Messages:
    9,733
    Likes Received:
    2,046
    GPU:
    Vega 64 XTX LiQuiD
    @mtrai
    Im using my Stable (now it's default) PP Table + WattMan for Fan curve (in OP)
    All is working perfectly.

    ==
    All i can say about new Adrenalin:
    Best OC
    Fastest (min. +10% in almost any game on average) ...
    ... and most efficient to date Temp vs mV vs tW
    Cool & Quiet thanks to WattMan Fan Curve.
    HBM2 OC is better.
     
    Last edited: Dec 15, 2018
    jimpsar likes this.
  15. slicer

    slicer Member Guru

    Messages:
    109
    Likes Received:
    34
    GPU:
    Gigabyte VEGA 64
    [​IMG]
    How would You start to test what mV is suitable for P1 etc?
    Do I need to set all states to one mhz level from P1-P7 same with mv under those states and just open up for example FireStrike in windowed mode and game2 in loop?
    Then monitor when it hits the default P1 speed slightly downvolting from default mv?
    Altough if I do it that way: set all P1-P7 states to 1138mhz and vcore to 900mv, with that settings it wont reach that mhz max clock or even average clock in FireStrike test method.
    I have to modify P2 states to higher ~1302mhz so I could get max P1 to show up. Why is that?

    It seems like P1 state is linked to P2 somehow.

    See picture, to exactly show what I mean.

    https://imgur.com/a/7cgBycY

    I set all the states to the first P1 value, also voltage. And ran the FS test, core only managed to hit 1121mhz (around -17mhz shy of P1 1138mhz). Only if I raised the P2, then it could reach P1 state (1137+mhz in GPUZ stats). But it is confusing, do I have to modify P2 to get P1 mhz in place?
     
    Last edited: Dec 15, 2018

  16. OnnA

    OnnA Ancient Guru

    Messages:
    9,733
    Likes Received:
    2,046
    GPU:
    Vega 64 XTX LiQuiD
    P0,P1 up to P3 Just a notch lower.
    Rest is easy to test, all you need is to play some non-heavy game.
    e.g. Up to 1200MHz can be tested with MtG Arena / The Elder Scrolls or similar card game.

    Don't test this low P-states in 3Dmark (it's too heavy)
     
  17. mtrai

    mtrai Maha Guru

    Messages:
    1,125
    Likes Received:
    332
    GPU:
    PowerColor RD Vega
    Firestrike With tessellation off using the Vega SoftPowerPlay Editor on PowerColor Red Devil Vega 64 flashed with a LC bios.

    Firestrike Graphics 29156

    [​IMG]
     
    OnnA likes this.
  18. OnnA

    OnnA Ancient Guru

    Messages:
    9,733
    Likes Received:
    2,046
    GPU:
    Vega 64 XTX LiQuiD
  19. OnnA

    OnnA Ancient Guru

    Messages:
    9,733
    Likes Received:
    2,046
    GPU:
    Vega 64 XTX LiQuiD
    @mtrai
    Yup, finally we are close to 30k

    -> Now test for me with Tess x8 (should be 29k IMO) and Default.
    I can see the difference in Overall & Combined ;)
    Min. 850W PSU is needed for this score.

    Difference between x8 and Default is ~100pts

    Note: Driver is not Yet approved (but it will be )
     
    Last edited: Dec 15, 2018
  20. TonyTurbo

    TonyTurbo Master Guru

    Messages:
    233
    Likes Received:
    16
    GPU:
    Vega 64 @ 1750/1180
    needs a valid 3dmark score not one with trickery in the control panel to cheat the score.
     
Thread Status:
Not open for further replies.

Share This Page