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. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,514
    Likes Received:
    2,912
    GPU:
    MSI 6800 "Vanilla"
    Wattman works but by default as Onna said it only covers the two last power states for the full 3D clock speeds which is alright for the most part but for fine tuning you can (For now at least.) also use this registry setup to allow for editing the others and have more control over voltage and clock speeds even during the lower draw states for the GPU. :)

    The last two should be for 3d and then boost with the result usually being that as the GPU hits full load it'll be somewhere in-between these two states using the settings you've modified via Wattman however p5 and lower comes into play for less demanding titles where the GPU will be at these levels instead because it doesn't require more and that's why editing these can see a further reduction of voltage and maybe modifying the clock speeds for power consumption.

    Still trying to read up more about Vega and also issues with Wattman and the current drivers but it's not easy to find anything else than what's been mentioned already though that's already pretty plentiful as is. :)

    I did recently learn that the power draw actually affects total and it's not just a "up to" value so a full 50% would actually negate a reduced voltage and modified clock speed negatively by having the card consume more power than needed.
    (I was expecting it to draw more as needed by having a higher total allowed before it would throttle and use that as required but no it seems it just has the GPU draw more power and that can adversely affect power consumption and thermals.)

    And then there's Chill and it's influence on things, balanced versus custom and more monitoring, in effect cutting the 300w or above down to 260w, 220w or even lower at 180w power usage or less.
    For Vega 64 the guaranteed Samsung HBM2 chips and being able to hit 1000+ Mhz on these is also helped with a higher core clock speed, for Vega 56 a bios flash is often required if compatible and then you can hit 1000+ on Samsung although on Hynix 850 or maybe 900 Mhz is likely going to be the cap and then it doesn't scale quite as much so it's possible to reduce clock speeds and only see a small drop in performance for a further decrease in power consumption.

    Although it varies from game to game, some demand more and others are going to be pretty lenient on the hardware so it's the usual with testing and tweaking. :)

    1500 Mhz at 850 Mhz for the HBM2 memory at 1.0v isn't too bad for what I've settled on so far, lower might be possible too but this is much better than the defaults and there's not really any significant difference in performance plus it's stable.
    (For the Vega 64 1700 at 1100 HBM2 is doable with some tuning or reducing it for a slight performance drop and a nice power consumption reduction and less thermal though V64 liquid isn't going to be too bothered by heat heh.)

    For the memory itself I think there's a pretty nice scaling up to at least 1000 Mhz or maybe 1050 and then it starts getting a bit more uncertain with what's needed for stability and how much of a return you get in terms of performance and how much the core clock speeds starts to play a bigger role but the cards certainly can scale very well though it does take some testing and tuning of the settings since every GPU is going to differ although at least Vega 64's remove the gamble with if it's going to be Samsung or Hynix HBM2 modules since memory overclocking can yield some really nice gains if possible.

    Though if not reducing the GPU power draw and clock speeds can be achieved for a very small performance hit instead of AMD's hefty 1.2v default though as Onna's various tests show Vega 64 also scales really nicely since it's often very high quality binned GPU chips in these. :)



    But yeah you can stick with Wattman or just the default power states if you want to and that's going to affect most of the 3D clock states during more recent games either the lower one or the higher one depending on total although it seems the GPU prefers to stick to something in-between both although closer to p7 if it's not thermally throttled or otherwise limited by various factors. :)
     
  2. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,514
    Likes Received:
    2,912
    GPU:
    MSI 6800 "Vanilla"
    Driver quirks is also a thing though, voltage spiking to max if you use downsampling was a issue and then there's some games putting a higher load on the card and it might not be entirely stable suddenly if it worked in all other titles and software tested so far, Wattman not fully corresponding to Vega's hardware and some of the settings being a bit vague as to what they affect for example though it's primarily the HBM2 memory voltage which is something else since HBM2 voltage is hard set to a single value and then fan speed only kicking in during load so it reports as 0 otherwise. :)

    Stability too and max possible settings could work better from one driver to another and then get worse, same as NVIDIA I believe but it tends to settle down after one or two more drivers and it could be related to numerous factors from just ambient temperature to something with the OS or the software tested too.

    EDIT: And that's too many smileys now ha ha, but yeah the usual I suppose with setting up values and then testing to verify stability and saving them. Ideally also saving the file as to have if the driver crashes or something causes Wattman to load the defaults again instead.
    (Fast boot is still a compatibility issue I believe with the current drivers and causing the GPU to revert to default speeds or well, Wattman will behave as if it crashed.)
     
  3. Slinkyminx

    Slinkyminx Master Guru

    Messages:
    209
    Likes Received:
    75
    GPU:
    GTX 1060
    I do really appreciate all that, thank you.

    For now I have just done this

    [​IMG]

    Went for lower clocks to try and get really low voltage, and 1000 seems happy on the memory.. Although after what you have said I am going to try and drop the power target..
     
  4. Dekaohtoura

    Dekaohtoura Master Guru

    Messages:
    345
    Likes Received:
    45
    GPU:
    Sapphire V56 Pulse
    Hmmmm...interesting reading material.

    My Pulse 56 (Samsung mem) can't go above 1560 core (P6 1562/1070mV, P7 1592/1090mV), while peaking at 1080mV (Heaven Valley), regardless of Power slider ("normal" @+20%, experimenting @+50%).

    Mem@930 (950 is stable, but I'm trying to keep the HBM below 75oC).

    Using Wattman, since OverdriveTool doesn't seem to change anything.

    Open to suggestions, as always.

    edit:
    There seems to be some hysteresis (offset) either at Wattman in general, or specifically at the Pulse model.

    Setting P6 @1602 and P7 @1652 (same voltage), P+50%, allows the core to max out @1613, using only 1063mV (Valley 1440p, 3dMark TimeSpy).

    Will keep testing...
     
    Last edited: Jul 28, 2018

  5. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
    WinX Pro Only Tweak.
    open cmd as Admin:
    Copy/Paste

    powercfg -duplicatescheme e9a42b02-d5df-448d-aa00-03f14749eb61

    Then go into Power Plan and Set this bad boy up.
    (also good to have this soft for better Tweak e.g. Min to 5-30% instead of 100%, which is sometimes useless)
    -> http://www.mediafire.com/file/wt37sbsejk7iepm/PowerSettingsExplorer.zip
    THX to Our member mbk1969

    You'll end up with Best possible Latency in Gaming etc.

    [​IMG]

    -> https://www.theverge.com/2018/2/15/17015674/microsoft-windows-10-ultimate-performance-mode
     
    Last edited: Jul 28, 2018
  6. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
  7. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
  8. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
  9. Synoxia

    Synoxia Member Guru

    Messages:
    148
    Likes Received:
    14
    GPU:
    Rog Vega 64
    Hi guys, what's the optimal HBM temp before it starts to underperform?
     
  10. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,514
    Likes Received:
    2,912
    GPU:
    MSI 6800 "Vanilla"
    65 degrees in Celsius for Vega as far as I'm aware, it then loosens the memory timings which sees a slight decrease in performance as a result. :)

    EDIT: For Fury and HBM1 though now that I check your GPU that's I'm not actually sure although I assume it has something similar though it's more static in it's profile for memory speed to where I believe it only has a single operating mode whereas Vega has three which also differ between Vega 56 and Vega 64 :)

    Or perhaps it's Hynix versus Samsung, well the states should be 167 Mhz idle, 500 Mhz low / 700 Mhz low for V.56 and then 900 Mhz high \ 800 Mhz high for v.56 which can then be overclocked a bit although not as much on Hynix memory modules.
    (Whereas for the AMD Fury it retains 500 Mhz constant though it has a wider bus width at 4096 instead of 2048 so still quite a bit of memory bandwidth for that GPU though it's not hitting the full theoretical limit of 512 GB/s but tends to be around 370 GB/s from what I remember of reading when this was tested since core clock speeds are holding this back.)
     

  11. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
    Depends on GPU but Save is up to 75 deg IMhO :D (60-75 deg. cels. is not much for todays H/W)
     
  12. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
    At 11min. he talks about why Vega Instinct 25 MI is getting better & better in Data Centers.
    And now they will have ready 7nm GPUs in few months.

     
  13. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
  14. Synoxia

    Synoxia Member Guru

    Messages:
    148
    Likes Received:
    14
    GPU:
    Rog Vega 64
    I've got vega now. so no point in overclocking HBM after reaching 65c, correct?
     
  15. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,514
    Likes Received:
    2,912
    GPU:
    MSI 6800 "Vanilla"
    If it hits 65 degrees on the HBM2 sensor it will drop the memory timings but I think it will try to keep the clock speed but it will still impact performance somewhat. :)
    ROG Vega 64 means the Strix model right? If so and if that's the first of the custom Vega cards then Asus might have taken a few shortcuts with thermal pads and having full contact on every component so it's the VRM sensors you should keep an eye on, I think these are rated to at least 120 degrees Celsius (Fury was 150) but you don't want them to get too hot and that will also heat up the card itself if these get really heated up.

    It varies from GPU to GPU though and I think they released a custom bios as well but it's not always playing with Wattman's settings going by the AMD reddit posts though it might be more for the fan speed rather than clocks or voltage settings so if that's so I assume they modified the fan curve a bit to keep the card cooled further by simply keeping the fan at a more aggressive and higher RPM value or having it spin up faster and starting from a lower temperature.

    Just something to be aware of, if the thermal sensor look fine for the VRM's chances are newer production GPU's might have been fixed too so no problems with these.
    (If it's the same issue then they either skipped on pads for the VRM's or they're not good enough or don't have full contact with the cooler, whichever it is that's why those get a bit hotter as a result sitting around 105 to 110 degrees for some GPU's.)


    EDIT: Well depending on the GPU workload and how much power it is drawing the VRM's do still get hot so 90 - 100 isn't entirely unreasonable when it's going full load either.

    EDIT: And a bit of reading later and so far things look good for the GPU, some of the programs for monitoring the sensors (HWInfo being listed specifically.) might be a bit off or pick the wrong sensor too so that might explain some reports. :)
    (Especially if it picks the sensor that's usually called the hot spot which will usually be a bit hotter than the others.)
     
    Last edited: Aug 2, 2018

  16. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
  17. Synoxia

    Synoxia Member Guru

    Messages:
    148
    Likes Received:
    14
    GPU:
    Rog Vega 64
    Did you flash Vega 64 LC bios on this right? How much power target does regular vega 64 require?
     
  18. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
    No, I've bought Original Vega 64 XTX LC (it's 1 of 800 pc.)
    Proud owner of Limited Edition :D

    I'm using Max of 1% Power or 0% For Gaming that is.
    For 3Dmark it's 25% or 12% depends on Bench.
     
  19. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
  20. OnnA

    OnnA Ancient Guru

    Messages:
    13,890
    Likes Received:
    3,644
    GPU:
    3080Ti VISION OC
Thread Status:
Not open for further replies.

Share This Page