OverdriveNTool - tool for AMD GPUs

Discussion in 'Videocards - AMD Radeon' started by tede, Aug 6, 2017.

  1. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,564
    Likes Received:
    2,961
    GPU:
    XFX 7900XTX M'310
    I wouldn't trust Windows Update to get the proper drivers so the only thing that should do is ensure the existing install isn't suddenly replaced by whatever the updater found and installed, seeing how it's a mix of mobile and desktop it might even try to download a separate driver for the Vega8 and I don't think that would go very well.

    Thinking of it I might have forgotten one important detail and that's if the Vega 8 shows up properly in the device manager it should have the driver installed and functional and it might just be the software that isn't fully compatible with it yet.
    (Or a limitation in the drivers perhaps, that could also be a possibility if Vega 8 is handled differently being a APU compared to what's reported and available for the dedicated GPU's which it does find.)

    Or other software like GPU-Z or AMD's control panel itself for that matter. Device manager would show current status though but if the driver installed fine that could be the cause, I might have been a bit too quick to jump on it as a driver issue. :)
    (For AMD's CNext software that would be under System and then there's separate tabs for overview, software and then a tab for the GPU's it detects.)
     
    Chung likes this.
  2. Chung

    Chung Guest

    Messages:
    6
    Likes Received:
    0
    GPU:
    ROG-STRIX-RXVEGA56-
    Yes the Vega 8 shows up on Device Manager. By AMD CNext do you mean a CNext.exe file ? I have no idea what that is :( Shall I just run it and see the GPU it detects?
    Also, since I don’t really want to OC the Vega 8, is it possible for OverDriveNTool to ignore the Vega 8 and only tune the Vega56 ?
     
  3. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,564
    Likes Received:
    2,961
    GPU:
    XFX 7900XTX M'310
    Ah that's Catalyst Next or Radeon Settings as it's called now. I keep mixing it up with how it was CNext.exe for a couple of drivers and then RadeonSettings.exe afterwards but yeah it's the software for AMD's settings which would have a tab for system showing info such as the software installed and then also the GPU's it would detect but if it's shown in the device manager it's working. :)

    And OverDriveN is per-GPU I believe so it should definitively be possible to ignore the Vega8, in fact if it doesn't see it but with it showing up in the device manager and thus working as it should it would be skipped over and shouldn't be a issue.
    Will have to try and look up more info on these on-board Vega models, could be how they work and not being part of Overdrive and thus not overclockable or at least not by the same method or perhaps the software can ignore APU's and focuses only on dedicated GPU models. I'd imagine the CPU on-board GPU would definitively handle a bit differently so it not showing up in the regular Wattman settings might be the reason why it's not detected if OverDriveN-tool is going by this for setting any changes.


    EDIT: Well it does overclock but not through standard means from what I can see, either bios or a separate software so that might be the reason it wouldn't show up if it's not going via Wattman and it's Overdrive (Overdriv N now.) API at all. :)
    (AMD Ryzen Master Utility, looks about the same as Wattman but for the CPU. https://www.amd.com/en/technologies/ryzen-master )
     
    Chung likes this.
  4. Chung

    Chung Guest

    Messages:
    6
    Likes Received:
    0
    GPU:
    ROG-STRIX-RXVEGA56-
    I've been messaging tede and he told me to go to setting and there is a setting to ignore unsupported GPUs. Since I only want to mine with the Vega56s, I turned it off. Now I'm able to create the Run_OverdriveNTool.bat file as in vegaminingguide.

    However, I'm hitting another problem. Running the Run_OverdriveNTool.bat seems to do nothing: the screen did not go black and the fan aren't spinning. I go into OverdriveNTool and it seems all the settings are applied (e.g. Fan minimum is set to 3000). Yet the fans aren't spinning at all.
     

  5. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,564
    Likes Received:
    2,961
    GPU:
    XFX 7900XTX M'310
    Does the utility support overriding Vega GPU's fan settings would be my first thought then.

    The GPU driver not re-initializing is also curious if it's meant to for some of these overrides but for Vega GPU's the fan speed settings don't seem to take effect in the unsupported lower power states so it would be p6 and p7 for the lower and full 3d clock speeds and bios editing is out due to restrictions in hardware on modding these.
    I know Afterburner and I guess also TriXX can force a higher speed for the GPU fan so it can be done but it's something I will have to look up.

    EDIT: No it doesn't seem like it does.
    https://forums.guru3d.com/threads/overdriventool-tool-for-amd-gpus.416116/page-18#post-5551399

    Or perhaps that's just for this parameter although I'm guessing it applies to min and max too because of course Vega has to be complicated.

    Eh it's not a bad GPU but it doesn't seem like Wattman is adjusted fully for these changes so some settings are outright wrong or certain overrides don't work fully.
    (HBM voltage being GPU floor voltage or that is it's the min voltage for one thing that stands out.)

    Refreshing in OverdriveN Tool should show if it accepted the other values at least, assuming the bat file is set up to apply to the correct GPU.

    And for Crossfire if it's the second, third or fourth GPU it might currently be using ULPS since it wouldn't be active during idle speeds so that might affect things a bit as well if it's almost in a disabled state which is often why you'll see recommendations on disabling ULPS in the registry. :)


    EDIT: There is still a lot I am not fully familiar with myself, Vega has been a learning experience but it's quite different in how certain things work and the driver likely still has some room for improvement even if some initial kinks have been worked out since launch. :)


    EDIT: How to explain, it's more like Vega is operating within a state of parameters or fluctuating constantly unlike previous models which had more of a absolute setting and operating parameters.


    If you were to set the GPU to operate around 1400 - 1450 for p6 and p7 chances are even with reduced voltage and a high fan curve eliminating both voltage and temperature as possible throttle areas well it's going to be in-between the two most of the time.

    Increase to 1550 - 1600 as per the stock settings for the V.56 and it'll be in-between these two instead as long as voltage is supplied and temps hold up so it certainly wouldn't have had any problems hitting p7 1450 in the earlier example constantly but it's just in-between instead most of the time from my own experiments so far.
    (This could very well be by design, min and max 3d clock speeds and the ideal is somewhere in-between for how the GPU operates with these settings.)

    Voltage is also very important both for temperature and the overall clock results but then it can also spike which sees a higher clock speed for a short period of time and this is when the GPU driver might either crash or reset if voltage is too low making it a little bit trickier to undervolt the GPU at least for gaming which some games or game engines are more susceptible to this but for a more constant even workload such as with applications it's usually less prone to this behavior.

    GPU-Z or similar would also confirm if it's Hynix or Samsung memory, V.56 without a bios flash would operate at 1.25v for HBM2 so overclocking is limited regardless but Samsung can hit 900 - 950 depending on GPU which as always it's going to vary whereas with Hynix some might be 850 to 900 or not overclock well at all before you get artifacts or memory errors, voltage via flashing might help but more so for the Samsung chips.
    (Early shipments even for V.56 used Samsung to even if Hynix is more common and almost used exclusively now. Tried submitting this GPU's bios to TechPowerUp and saw they already had a bios but that was with Samsung and this is Hynix so that must have been a first batch or review model.)


    And the fan speed, wouldn't mind a higher minimum but that's going to require either AMD updating their Wattman API or using a program like Afterburner to set or force a manual fan curve or temperature limit overriding AMD's drivers it would seem.
    (It works pretty well at full load at least, keep a program like GPU-Z for checking other sensor temperatures for the hot spot and HBM2 and keeping the card around 65 - 75 degrees would be best but depending on load and what it's working with this can be up to 80 or higher where throttling will start affecting the card more.)


    EDIT: And of course Vega.56 using the chips that didn't become Vega.64 is using pretty high defaults for voltage ensuring they operate without issue at stock speeds but this also allows for some pretty decent under volt results with some tweaking and testing to ensure it's completely stable. :)
    (It might not be a V.64 especially the liquid models but it can still perform really well at less voltage and without losing performance or at least far less performance than how much you can reduce voltage usage on some cards.)


    Well I guess it's fairly evident from just trying to adjust some of the settings in Wattman either directly or via utilities such as this one and seeing the results from that.

    Though I do wonder if AMD would tweak the interface a bit to make it more clear what the settings do and what they can affect and in which situations.
     
    Last edited: Jun 3, 2018
    Chung likes this.
  6. OnnA

    OnnA Ancient Guru

    Messages:
    17,793
    Likes Received:
    6,691
    GPU:
    TiTan RTX Ampere UV
    I'd like to have Set Fan for my Fiji also :D
    Now im on MSI AB (33% for daily use, 52% for Gaming)
     
  7. MaxMidnite

    MaxMidnite Member

    Messages:
    11
    Likes Received:
    0
    GPU:
    ATI RX580
    Is there a way to add new feature so it detects the memory make like GPUZ.
    I often get confused which GPU is what memory and each memory produces different overclocks hence the need for this.??
    Is this possible to do or will require lot of work, I worried that the GPUZ drop down does not correspond to ONTool one.??
     
  8. faridgv

    faridgv Guest

    Messages:
    7
    Likes Received:
    1
    GPU:
    rx 470
    @tede is there any way to set gpu core&memory voltage under 800mV??why overdrive show error code :-1?
     
  9. DrathVader

    DrathVader Guest

    Messages:
    3
    Likes Received:
    0
    GPU:
    RX 480 Nitro 4GB
    Just tested it, power efficiency works flawlessly. Thanks mate, I can finally ditch Wattman completely.
     
  10. tede

    tede Member Guru

    Messages:
    144
    Likes Received:
    71
    GPU:
    AMD RX 470
    @OnnA try 0.2.7beta4, setconstantfanspeed uses now also Overdrive6 as well as Overdrive5 API, I don't know what is proper for Fiji, maybe none of them work at all.
    Overdrive5/Overdrive6 is an older API that was used before Polaris (Fiji?) to set fans or OC.
    From what I found it still works for Polaris and setting this to manual % speed will override OverdriveN API fan settings, which means Wattman or OverdriveNTool normal settings are ignored and not used, untill you set setconstantfanspeed back to Automode.
    What is the problem with GPU-Z? It has the same GPU order. Alternatively HWiNFO64 also displays memory type and comparing PCI Location with OverdriveNTool bus number you can be sure what gpu_id uses what memory type.
    This is low limit for your GPU, you can't normally set voltage below limit in OverdriveNTool, from what I know you have 2 options:
    1. Use I2C offset, if you can't use it in OverdriveNTool, use other programs that set voltages using I2C like Afterburner
    2. Edit voltages in bios or SoftPowerplayTable, but later you won't be able to change clocks/voltages in OverdriveNTool because you will be below limit, which result in error -1.
     

  11. OnnA

    OnnA Ancient Guru

    Messages:
    17,793
    Likes Received:
    6,691
    GPU:
    TiTan RTX Ampere UV
    setconstantfanspeed ? where?
     
  12. tede

    tede Member Guru

    Messages:
    144
    Likes Received:
    71
    GPU:
    AMD RX 470
    Nevermind, I thought you was talking about this :)
     
  13. faridgv

    faridgv Guest

    Messages:
    7
    Likes Received:
    1
    GPU:
    rx 470
    @tede when a card have 750mV in default bios thats possible to set voltage under 800mV but when it have 800mV in bios I cant set it below 800mV
    i have some msi rx470-miner and now it working stable with 800mV with power consumption about 40-43watt but i want if it possible set this limit below 800mV
    i can not use I2C
    sry for english:oops::oops:
     
  14. OnnA

    OnnA Ancient Guru

    Messages:
    17,793
    Likes Received:
    6,691
    GPU:
    TiTan RTX Ampere UV
    Is this true that Vega XTX LC Version comes with additional P-State 8? (0-8 not 0-7)
     
  15. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,564
    Likes Received:
    2,961
    GPU:
    XFX 7900XTX M'310
    I think they do at least, 1700Mhz for the last state instead of 1663 but I'll have to look it up.
    It could be however that as you mention it's 0 - 7 and my initial info was wrong because I assumed it had 8 states and then the zero state.

    I'll have to re-check that now. They all differ a bit. :) Frontier, 56, 64, 64 LE and 64 water/liquid.

    EDIT: Yep, it's 1536 for the 64, 1663 for the liquid but it's the same amount of states so it has no extra state.
     
    OnnA likes this.

  16. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,564
    Likes Received:
    2,961
    GPU:
    XFX 7900XTX M'310
    And the total power draw is quite noticeable for a very small bump in clock speeds.
    https://forums.overclockers.co.uk/t...king-undervolting-information-guide.18793012/

    So while it's an overclock it's not that different from the stock configuration so roughly hitting 1615 Mhz at 1100mv up to 1640 Mhz at 1200mv and it changes going from 260w to 320w
    Small wonder downvolting these are so popular, you would lose little to nothing for a notable power draw decrease and thus less heat stabilizing the clock speeds in the process.
    (The actual clock speed input for p6 and p7 is different but Vega tends to be in-between the two even when conditions allow for full p7 speeds from my own testing down at 1300 - 1400 Mhz where there's no real issue with thermals or power.)

    They wouldn't have competed head on with the 1080 at 1450 - 1500Mhz and maybe up 1550 - 1600 for the higher-end versions but it would have made them a lot more efficient instead of trying to squeeze out what? Barely 2 - maybe 3% extra performance perhaps maxing the card out like that?

    That and hitting 1 Ghz for HBM2 speeds instead and you'd still have a really good product, sadly pricing would still be a issue and the mining popularity and lower shipment availability would still have happened. Perhaps even more so if miners didn't need to re-adjust the GPU on their own any more.

    After all, 100mv decrease is quite serious and then it barely hits performance at all and the GPU overall actually responds better maintaining the higher clock speeds and perhaps not throttling memory due to the heat.
    Though I guess the Vega videos in the other thread pretty much confirms that too, even the 64 with it's additional shader cores and higher clock speed can be taken down a notch with some careful tuning and testing and the result is overall almost a unnoticeable impact on performance for much better thermals, power draw and keeping the GPU more stable rather than hitting throttling all the time. :)


    EDIT: Which this utility can then help with by setting up more fine grained control over various parameters. Card certainly benefits nicely from this sort of fine tuning.

    EDIT: Which well it'll differ from one GPU to another just what the limits are but with some tuning and tweaking and testing in a few different games and benchmarks stressing the GPU well 1500 - 1550 Mhz seems like a good range and then scaling drops a bit or at least in the above tests it did, it's fully possible other titles will see varying results but it's a good initial overview of clock speeds, voltage and resulting power total and finally the performance from this whether it's a gain or a loss and how much.

    I'm still learning though, plenty of user reports and full on guides or even entire videos though so plenty of material available showing well it's pretty similar results usually, AMD's stock settings push the GPU pretty hard for a small bump in performance but toned down it's still really competitive and the drop isn't that big for overall performance compared to how much you can tune the voltage and resulting wattage. :D

    Or how to say, no secret that both Vega and the earlier Fury GPU's ran quite hot and quite high for AMD's designated target values but this also allowed for some nicer fine tuning.
    (-25 to -50mv for Fury wasn't bad really, full on -100mv here though that's really good.)
     
    Last edited: Jun 21, 2018
    OnnA likes this.
  17. Chung

    Chung Guest

    Messages:
    6
    Likes Received:
    0
    GPU:
    ROG-STRIX-RXVEGA56-
    Might sound like a stupid question, but are the voltage toggles for the core and memory a detrimental setting? What I mean is, if I have the voltage for the core set to 950mv, does this mean the core will always get 950mv, or does this mean its get 950mv at most? Also, can I fry my card by supplying too much voltage? Assuming core/memory clock unchanged.
     
  18. OnnA

    OnnA Ancient Guru

    Messages:
    17,793
    Likes Received:
    6,691
    GPU:
    TiTan RTX Ampere UV
    Hmm, my Vega shows 1750MHz 1.250v as Default :eek: lol That V
    So we need to UV this Bad Boy :p

    Edit- UPD.

    Yup Jonas
    -> HBM_2 v Readings in HWinfo = 1.356v (But iMO it's HBM/HBCC controler, the actual HBM_2 voltage is regulated in house with Wattman or OverdriveN Tool)
     
    Last edited: Jun 27, 2018
  19. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    17,564
    Likes Received:
    2,961
    GPU:
    XFX 7900XTX M'310
    Memory voltage is hard set to 1.250mv for the Vega 56 and 1.350mv for the Vega 64, depending on memory type the timings and characteristics might differ a bit though most Vega 56's do seem to ship with Hynix now though Samsung is still used though it's prioritized for the Vega 64 model.

    For Wattman it's either a lower power memory state or it's the floor voltage and possibly related to other voltage states but setting this to e.g 1.0 seems to force it so the GPU never goes below this value for the core voltage.
    There's some more research on the bios and certain settings found here: https://www.overclock.net/forum/67-amd-ati/1633446-preliminary-view-amd-vega-bios.html :)

    And yeah undervolting the GPU can bring some really nice yields, full on liquid Vega 64 probably handles the thermals regardless but it doesn't hurt to try and find a stable lower voltage, -50 or even -100mv seems doable on many of these GPU's though exact limits will vary and some models are less capable than others. Gigabyte seems to be having issues going by recent posts on the AMD Reddit for example. https://www.reddit.com/r/Amd/comments/8u60ze/buyer_beware_gigabyte_radeon_rx_vega56_gaming_oc/

    Whereas say Sapphire is usually considered pretty high end although being a AMD exclusive partner they might have some preferential treatment and early access to chips and third party models for PCB designs including the initial Vega Nano variant now used on their Pulse models though other manufacturers now offer similar models for Vega 56 with Vega 64 rumored for later though we'll see. :)
    (Nano is doing alright but has less power phases at 7+1 instead of 12+1 although still capable and still very well engineered but with bits being a closer thermals become more of a factor thus why Pulse and Red Dragon have these extended coolers.)


    EDIT: The reference models should be alright regardless of manufacturer of course, they're all the same technically after all.
     
    OnnA likes this.
  20. Draka Joko

    Draka Joko Guest

    Messages:
    13
    Likes Received:
    0
    GPU:
    radeon 570
    Hi.
    When setting the OC clock on amd nitro+ 570 to 2250 in overdriventool it doesnt work. I can only go to 2225.
    What is causing this?
     

Share This Page