RDNA2 RX6000 Series Owners Thread, Tests, Mods, BIOS & Tweaks !

Discussion in 'Videocards - AMD Radeon' started by OnnA, Oct 29, 2020.

  1. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    I get this "The previous overclock settings have failed, system has been restored to it's default settings. Press F1 to enter setup. Press F2 to load default values and continue." CPU overclock related message albeit not always which does point to CPU being the culprit. Don't have a Ryzen CPU at the moment just a lowly i7 8700 non-k. Also, I have SUA 2200I ups so no issues with brownout, power spikes etc.
     
    JonasBeckman likes this.
  2. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,445
    Likes Received:
    2,813
    GPU:
    MSI 6800 "Vanilla"
    Yeah standard bios error message I had every time there was some work done on the sewage pipes or a fallen tree or anything taking out the power cabling I just see it as a generic bios message informing the user the settings reverted to default and a quick load of the saved profile fixes it. :)

    EDIT: I had almost the same thing on the X79 system and also here with the X570 board when this happens.

    And on to doing some reading on 3DMark and Port Royal.

    Looks like a fairly standard D3D12 benchmark should be a good stress tester.
    But what I am looking for is how it affects CPU power draw if we're looking into undervolting and instability from this.

    Intel board though so my experience on that even though it was a older 3930k was that it just ticked along and could handle almost anything being nowhere near sensitive like the Zen2 3900X I'm on now is.

    Different approaches and solutions though.
    For the 8000 series CPU's and on a thing with Intel refreshing and updating 14nm over and over is that it becomes really good and mature. :)

    I would expect it to be incredibly robust as a result.
     
    Last edited: Jan 1, 2021
  3. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,445
    Likes Received:
    2,813
    GPU:
    MSI 6800 "Vanilla"
    Wonder how the Intel Turbo frequency and boosting is kicking in here.
    You have the TDP and what Intel sets but then you also have the boost power draw and requirements to maintain this unless you lower the frequency, non-K series so a locked multiplier but it still works very similar.
     
  4. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    I have reset my CPU to Optimized defaults, I guess today's full night run will tell me if my system is stable or not with these extreme settings. If not, then I will try with default settings, it's going to be a few busy days debugging this. Just FYI, My card can almost reach a 3080 in port royal ray tracing test (https://www.3dmark.com/pr/724618) and given the low OC potential of Ampere GPU's, I think I made the right choice going with RX 6900 XT, if these settings hold up. I do have a 2080ti lying around which I can use for PSU testing but it won't draw good amount of power.
     
    JonasBeckman likes this.

  5. Shadowxaero

    Shadowxaero Master Guru

    Messages:
    205
    Likes Received:
    33
    GPU:
    Red Devil 6800 XT
    Sigh I been trying to push my card more, but I think I will need a 5000 series CPU....and I can't for the life of me get my hands on a 5950x. I wish I could increase just a little bit to 1162 but the driver crashes on boot whenever I increase voltage with MPT :(
     
  6. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,445
    Likes Received:
    2,813
    GPU:
    MSI 6800 "Vanilla"
    Well I forgot a important part about Intel and the way the CPU scales and clocks under load. :)

    The CPU is rated at 65W but actual usage can hit or even exceed 100W so a undervolt might have been hitting instability from how Intel and AMD reports and calculates CPU TDP and the actual real value you get through monitoring it doesn't have to be a worst conditions stress test it can still peak at pretty high values compared to the expected.

    https://linustechtips.com/topic/936752-looking-for-8700-non-k-real-life-wattage-and-temps/

    [​IMG]

    Offset mode can be effective though but you have to test incremental smaller decreases making sure it's stable before continuing lower.

    And then there's Multi Core Enhancement or MCE as another factor but my knowledge about this is effectively null so I can't really say anything useful about it other than finding that it can further interfere with setting up voltage values. :)

    https://linustechtips.com/topic/874616-voltage-settings-on-8700-non-k/
     
    Last edited: Jan 1, 2021
  7. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    Hoping this is the case and my PSU and GPU are good.
     
    JonasBeckman likes this.
  8. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,445
    Likes Received:
    2,813
    GPU:
    MSI 6800 "Vanilla"
    Yeah I'm curious as to how the over current protection managed to get tripped by a few reviewers but overall the power draw and usage looks really fine on the 7nm+ node and the 6000 series even the 6900XT is pretty good using many of the same settings as the cut 6800XT and still being capable of scaling fairly well.

    Looks like there's a few very important factors for newer Intel CPU's and enhancements that I had missed which could very well be related to hitting instabilities when trying to lower CPU power usage either from manual voltage and settings or offset mode and a negative multiplier to lessen usage this way. :)
     
  9. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    You should ask @xcx xcxvgyt for his settings. He has his 6800XT tuned almost to perfection.
     
  10. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    So occt psu test completed with completed properly message for 1 hour but crashed the pc as soon as I moved the mouse. Running it again and this time will wait a minute after the test is completed before accessing pc.

    Edit : Forgot to mentioned that Intel XTU was installed when mouse movement crashed PC.
    Uninstalled Intel XTU, ran PSU OCCT test for one hour and passed successfully. Does this at least rule out PSU as the problem?

    Further Edit :

    3d mark custom port royal ran for 6 and half hours and failed with unknown error and not any hard crash. Stress test after this passed for port royal.
     
    Last edited: Jan 2, 2021

  11. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,445
    Likes Received:
    2,813
    GPU:
    MSI 6800 "Vanilla"
    It certainly doesn't seem like the PSU has any problems if you can run the tests for hours and it crashes after as the software is unloading it's not a hardware problem or at least not a PSU hardware problem it's been running at what the system requires for some extensive benchmarks usually pushing the components to their max so I would certainly say it's stable.

    It's hard to test these things though OCCT is a D3D11 derivative test suite so I usually go with Vulkan or DirectX 12 since it's more prone to hardware sensitivity and any one issue is useful to consider although it's hard to track down the source of the problem.
    OCCT's also been unable to properly stress the 6800 GPU I'm using it settles at 2.0 to 2.1 then in games it pushes the full 2350 even 2450 though this last one needs more testing to verify it's actually stable.

    Assassin's Creed Valhalla for it's VRAM and system RAM crash prone nature for any one error although you have to go between testing without the XMP and testing with the GPU at default or at least lower clocked ruling out which one it is if you get the game to reliably crash within 5 - 10 minutes is usually all it needs.

    Watch Dogs Legion with ray tracing to maximum (E3_Marketing instead of PC_4 preset.) works well too but even without this the card is working pretty hard for the game and it's similarly just a few minutes and it will crash usually more about GPU VRAM stability though than anything else.
    The ray tracing compared to Valhalla alters the GPU workload makes it shift a bit good for further testing although I kinda dislike how rough the reflections look ha ha. :p


    Think a actual protection shut-down from the PSU would also make it shut down for a period of time before it can be started up or it restarts on it's own although there's a whole number of methods on newer power supplies and my hardware knowledge is quite poor here.
    (I am really really impressed by what some of the higher end supplies are capable of but things like the Corsair 1600I are super expensive and way over the top for most users fine as it is or models just under or matching it's quality.)


    EDIT: For OCCT for compatibility I tend to disable the extra sensors all I need is for the GPU to pass a 5 minute test usually the first error (And there's only going to be one valid result, zero.) happens within a minute and then I know it's not stable.

    Test itself is not 100% but it's a good initial stress and stability test although I've only utilized it for the GPU specifically, complexity of 6 or 8 usually just max it and see how a 5 minute run works stopping on the first error found. :)


    5700XT was very sensitive so this pixel color detection I think is how it works would find hundreds and then thousands of errors once the clock speeds were above the supported and then the GPU's maximum even with more voltage and power.

    That card though was going with a undervolt since it's a 2% - 3% performance drop for a respectable lowered GPU junction and surface temperature readout and overall power draw can be lowered a bit too.


    6800 here can get close to a 6 - 8% performance boost in some titles though I'm keeping it a bit below it's maximum making sure it's holding stable without any major adjustments other than a small drop in voltage to slightly reduction junction and surface temperature since the dual slot cooler otherwise gets a bit hotter than the 2.5 slots the 6800 XT or 6900 XT's are using on the reference models.
    (Can drop the fans a bit too they get loud around 60% but they also do a relatively good job but at a higher noise level.)

    There's a bit more to gain and also more to lose from undervolting or clocking the GPU lower so a slight boost to clocks and also a bit of a boost to performance this time around instead of a minimal change. :D


    Same for the 6800XT and 6900XT probably even more so for the 6900 since it hits the default limits a bit more with the additional cores and while these help the clock speeds appear to be better than the GPU's overall scaling so they are more important.

    They all clock well though so the 6800 can get close to 6800XT stock performance but this card in turn can also gain close to 6900XT performance numbers and bridging that ~10% difference but then the 6900XT can also boost up and gain an additional bit of speed from running at similar levels 2.6 to 2.8 Ghz without further tweaks and having the additional cores.


    Fun although AMD capped them a bit so it'll be interesting to see what the actual limits could be if this can be lifted more than just some of the parameters.
    (These help stabilize the existing clock limits though so there's already a improvement from that alone.)



    EDIT: Now what did I miss.

    Oh OCCT and these sensors and compatibility.

    (From the FAQ on their website.)

    The little wrench icon and settings and for toggling reading from these.
    Those crashes are curious but I can think of trying without Corsair and EC (Embedded controllers.) unticked seeing if maybe that stops it.

    I generally untick them all since I do a pretty quick GPU 3D error detection run only not a full system test most of the time so all that's required is getting one single error or worse and then re-testing with lowered clocks until it's zero. :)
    (Can also tick the setting to stop after hitting a error since the ideal here is zero.)


    Usually works just doing a shorter 5 minute test and then if that holds I continue testing in various games seeing how these then handle the settings after that first run through OCCT.
    Thankfully many of the DirectX 12 games are pretty prone to instability from components not being quite stable even if D3D11 or various benchmarks can pass even longer runs.

    Port Royal and other D3D or Vulkan test suites should also be similar though AMD and NVIDIA might have more strict GPU profiles here, should still be much less than on stuff like Furmark which can almost throttle the card instead.



    EDIT: OCCT isn't the problem though but it's more what is causing it to crash like this and taking the system with it and what this might do with other software, running processes or monitoring utilities and overlays. :)

    Thus the focus on that and what might be problematic here and by extent if it's also going to be problematic with other software.
     
    Last edited: Jan 2, 2021
  12. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    Fixed by switching mouse and keyboard to USB 3.0 ports.
     
  13. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    Lots to digest so lets go one by one.
    Yeah I think that as well. Haven't had one crash since using Optimized Default settings in BIOS with XMP enabled and uninstalling Intel XTU. I have gone beyond stress tests and reached into the realm of torture tests and still PSU has not shut down.

    1. So what would would you suggest for testing through Vulkan API?
    2. AC: Valhalla and WD: Legions are crappy games and I don't wanna pay Ubisoft for crap even if they are good for testing. Any other alternatives in mind? How about metro Exodus?
    3. DX12, I am pretty much covered with advanced edition of 3dmark with all the upgrades.

    Good to know your experience with 6800. OCCT did not crash for around 5 minus for rx 6900 with max clocks as clocks are pulled downwards due to high TGP use, I think.

    Again, good info.
     
  14. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    So another unique quirk of these GPUs, fans don't go above 92% no matter what you do. Can someone else test it as well?
     
  15. PrEzi

    PrEzi Master Guru

    Messages:
    660
    Likes Received:
    529
    GPU:
    AMD 6900XT
    Yep, confirmed on my 6900XT. Goes to 3000 RPM (90-91%) when set to 100. Not a big issue though.... but they should adjust the ratio to match.... e.g. 100% should be the 3000rpm (and "old" 91%)
     

  16. OnnA

    OnnA Ancient Guru

    Messages:
    12,825
    Likes Received:
    3,202
    GPU:
    Vega XTX LiQuiD
    Vega 56 Vs 5700XT Vs 6800 :p

     
  17. MrSleezeBag1964

    MrSleezeBag1964 Member

    Messages:
    29
    Likes Received:
    3
    GPU:
    Radeon 6800 16 GB
    I figured the OC'ers might know, but I'm fighting a strange problem on my Asrock Phantom 6800. I Get random shut downs in normal, non OC'd use, but I'm having trouble replicating them. While I think it may be a PSU problem, even during shutdowns, the PSU doesn't fully turn off, but I'm pretty sure it only kills power to the GPU. Is there a forum-favorite logging software that may capture some info to help in my troubleshooting?
     
  18. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    Read some of the earlier posts in this thread. I has almost the same problem which turned out to be (at least no repeats since then) CPU voltages not being correct. Loading optimized defaults in bios resolved them. To confirm if it's a PSU issue, just download and run OCCT PSU test, it should croak within few minutes if it's a PSU issue. OCCT itself will provide you logging during the test, though don't know if it saves those logs.
     
  19. Godhand007

    Godhand007 Member

    Messages:
    46
    Likes Received:
    17
    GPU:
    RX 6900 XT
    So hang on, is the MAX RPM suppose to be 3000? According to MPT it's 3300.

    [​IMG]
     
    JonasBeckman likes this.
  20. PrEzi

    PrEzi Master Guru

    Messages:
    660
    Likes Received:
    529
    GPU:
    AMD 6900XT
    Good point !
    So either there is a bug (in the BIOS/fan control) or there is a bug (in the software) or both :D
     

Share This Page