Sapphire VEGA64 Nitro+

Discussion in 'Videocards - AMD Radeon' started by bernek, Jun 4, 2019.

  1. bernek

    bernek Maha Guru

    Messages:
    1,471
    Likes Received:
    33
    GPU:
    VEGA64 / 8GB
    Hey guys !

    Just switched again from green team (1070GTX) to red team and got a Sapphire VEGA64 ...

    It has a bios switch but I'm not sure about positions ... It was toward the front of the case and GPU-Z said 1630 MHz and I powered down and switched to the back of the case and the frequency in GPU-Z was the same 1630 MHz so my humble question is which one is the silent version ?

    Thanks in advance !
     
    BlackZero and OnnA like this.
  2. OnnA

    OnnA Ancient Guru

    Messages:
    9,723
    Likes Received:
    2,042
    GPU:
    Vega 64 XTX LiQuiD
    Congratulations on Nitro+ :D
    Update Your RIG.

    And in GPU-Z check is this 220tW or 264tW BIOS
    Look in Vega Thread OP....
    Welcome to the Club, HBCC always ON (HBM2 Cache)
     
    bernek likes this.
  3. bernek

    bernek Maha Guru

    Messages:
    1,471
    Likes Received:
    33
    GPU:
    VEGA64 / 8GB
    Thank you ! I've checked and both showed 264W :S

    I didn't have HBCC always on :(( will toggle asap !

    Thank you OnnA.

    Does it look ok like this ? or should I max the slider ???

    [​IMG]
     
    OnnA likes this.
  4. OnnA

    OnnA Ancient Guru

    Messages:
    9,723
    Likes Received:
    2,042
    GPU:
    Vega 64 XTX LiQuiD

  5. bernek

    bernek Maha Guru

    Messages:
    1,471
    Likes Received:
    33
    GPU:
    VEGA64 / 8GB
    Does this look better ? Monitor is 2560x1440 at 75Hz

    [​IMG]
     
  6. bernek

    bernek Maha Guru

    Messages:
    1,471
    Likes Received:
    33
    GPU:
    VEGA64 / 8GB
    What is the best way to OC these cards ? Undervolt by -25 to -50 mV then apply some clocks ? what about power target ? I'm just curious how good this GPU is but I've watched a few YT videos saying that undervolting is the way to go ?
     
  7. Exodite

    Exodite Ancient Guru

    Messages:
    1,913
    Likes Received:
    171
    GPU:
    Sapphire Vega 56
    There's a rather extensive thread on the subject that's worth a read.

    Generally speaking you can UV far more than that on most cards, at stock clocks I'd start testing with P7 @ ~1050mV and go from there.

    Each card is different and if you want to get the absolute max out of your card, or just enjoy tweaking in general, it's going to take a while to find the ideal balance of voltage, power target and clock states that balances everything out.

    Do give the linked thread a read. :)
     
  8. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    15,719
    Likes Received:
    1,712
    GPU:
    Sapphire 5700XT P.
    For undervolting if you aren't discouraged by a crash or two the automated undervolt functionality is fairly solid now and should try a value and then either the game shows errors or crashes and the driver will pick a slightly different value next time which you can monitor to find a baseline value and either add a +10mv to for stability or find a just low enough before it hits artifacts or stability problems maybe giving a slightly larger decrease.

    Upping the power limit can also help but will see the GPU draw more which also translates into heat even if the GPU core isn't taking as much voltage which if balanced together can lead to a decrease of the power throttling but also has to account for the thermal throttle for heat and the HBM throttle for heat which is around 70 degrees Celsius I think it was and then it loosens up the timings which for a Vega 64 GPU with confirmed Samsung you have a higher base clock and potential HBM2 overclock for 1050 - 1100 Mhz increasing by 20 or so and testing until either artifacts and stability and dropping down a bit though every GPU is going to have different characteristics and maximum values for memory and core clocks and required voltage but yeah I think the timings were a bit higher on Vega64 compared to Vega56 but the faster speeds should offset that and you have the Level1 memory timings setting in Wattman too for trying to tighten them just a little bit.

    It's a balancing act which I suppose is nothing new for overclocking but the GPU has various power draw and heat limits and tends to try and boost to around between P6 and P7 clock speed states closer to but not always fully at P7 levels so that fluctuation can also require some testing if it spikes to a bit higher which could be above previous stable test levels but it's not too hard to get a nice reduction in voltage in with a bit of testing and confirming stability either from the automated tools or lowering P7 and maybe P6 states gradually in small -10 or -20 increments and re-testing for any signs of artifacts or crashes.

    I'm on a Vega 56 here and the Pulse model which is a bit below the Nitro cards but still a fairly solid Vega card and hitting 1000mv for p6 now and 1050mv for p7 at 1500 and 1550 respectively which it sits around 1520 - 1530 most of the time. Could push to 1600 and Vega64 and the Nitro might go even higher and be able to sustain it though in my own case the HBM2 chips are Hynix with a slightly lower speed and overclock potential so going further sees very diminishing returns of 1% to 2% or so.
    (In fact the default is 1596 Mhz for P7 at 1200mv so I actually lowered it down a bit, tried 1500 too and 1450 for P6 but at 1400 it starts hitting performance a bit more noticeably though still less than I would have expected at maybe a 4 - 5 percent drop.)

    Higher is better and getting a lower core voltage and balancing the power draw can keep the throttling down or out entirely allowing for a sustained high clock speeds for GPU and memory and overall better performance but it takes a bit of testing though Wattman has a lot of options now for how to go about it and changes since Vega came out has seen improvements to how this works and how well the GPU handles and manages these various states and even the automated undervolt and overclock utilities can be a good way to find a baseline level to work with to either increase it a bit for undervolting and ensuring full stability for most any case or maybe a even lower level with some additional testing shaving off anywhere from 30 - 50mv more before hitting that final value. :)

    I'm pretty much a amateur and well it comes down to lowering it a little bit at a time for voltage and re-testing mostly, my own experiences are for this model of Vega and this particular GPU and each one will hit different values both for 56 and 64 although some characteristics can and will differ between the two. :)


    EDIT: It's nothing particularly new but the GPU's way to boost or fluctuate and the various (numerous) sensors and ways it can limit performance makes for a bit of tuning and testing to find the ideal values but it's not too complicated either whether you leave some headroom for a decent undervolt and possible overclock or try to find absolute values for how low it can go without issues with or without overclocking it further.

    Vega 64 is already clocked a bit higher too but with memory it can see some gains still although percentage wise outside of benchmarks it might not be very large increases though there's some potential for improvements so it's not like overclocking is entirely discouraged or anything.
    (Driver updates can also change things here compared to early tests showing a 2 - 3% GPU core increase and 5 - 7% memory increase for total performance and gains.)
     
    bernek likes this.
  9. OnnA

    OnnA Ancient Guru

    Messages:
    9,723
    Likes Received:
    2,042
    GPU:
    Vega 64 XTX LiQuiD
    Yup, it's fine.
    HBCC set to 12GB (move slider notch further)
    Launch game & see 12GB VRAM by HBCC.

    Thanks to this no hickups when commited >8GB :p
     
  10. BlackZero

    BlackZero Ancient Guru

    Messages:
    8,880
    Likes Received:
    470
    GPU:
    RX Vega
    You should always limit power usage.

    Higher power limits create heat, which again results in increased power usage, and you gain nothing.
     

  11. bernek

    bernek Maha Guru

    Messages:
    1,471
    Likes Received:
    33
    GPU:
    VEGA64 / 8GB
    My results have been far from spectacular ... I could only make it run worse or ran into some temp/power limitations ... So far AMD Wattman Turbo setting can beat my overclock easily ...

    So let me try elaborate what I did:

    1. Core clock on highest p state (7 I think) was 1660-1670
    2. Votage for GPU was -50 mW 1150 (tried also 1100)
    3. Power increase was from 0% - 50% (the most "good" one was around 20-30%)
    4. HMB frequency from 945 to 1100 (strange results in between). What frequency would I want ?! (that makes a point to OC to it)
    5. HBM voltage now here I'm lost it was shown 1100 mW then I set it to 1000 mV I can't have it on auto since its linked with GPU

    Steps 4 and 5 were the most problematic. I dont want more than 512 GB/s (1000 MHz HBM) its more than enough I think.

    Sadly either I'm getting old or this Sapphire Nitro+ is already at the max it can do. It is a HUGE card compared to my former 1070 GTX ... it looks like brick !!!

    What is the best way to test GPU clocks and temps these days ? I was using Unigine Superposition on EXTREME 1080p but I think that load is not realistic and thats why I'm running into limitations ???

    OnnA shared a few settings with me that were actually working but eventually the "magically" stopped working and card was going over 1600 MHz ...

    These were the settings:

    Best is to have 1.094v with 1668MHz and HBM2 at ~1080->1120MHz 943mV up to 975mV.
    Pow at +1% up to +12% (or with -5% for 'light' games)

    Is Wattman still bugged ? :D
     
  12. bernek

    bernek Maha Guru

    Messages:
    1,471
    Likes Received:
    33
    GPU:
    VEGA64 / 8GB
    I've found this article: https://www.Disney.net/guides/2990-vega-frontier-edition-undervolt-benchmarks-improve-performance

    And I've experienced the HBM downclock to 800 MHz ... so is this a bug ?! I only "play" with the higher P STATES and voltages ? Here is wher I'm wrong ?

    I've changed p6 and p5 also to not go higher than p7 voltage ...

    P.S. another issue that I have is that I'm not sure about the bios switch ... both positions show the same max rated tdp 264W ... I can't believe that this is not shown in the documentation or box of the card. So simple just write towards end of the case bios OC towards front bios is Silent version ... I'm not sure which is which ....
     
    Last edited: Jun 7, 2019
    JonasBeckman likes this.
  13. OnnA

    OnnA Ancient Guru

    Messages:
    9,723
    Likes Received:
    2,042
    GPU:
    Vega 64 XTX LiQuiD
  14. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    15,719
    Likes Received:
    1,712
    GPU:
    Sapphire 5700XT P.
    Interesting, suppose there might be more issues in Wattman than I was expecting although I am not sure on the P states since AMD unlocked them all.
    Bios switch can be a pain as well depending on if it's a overclock type with higher voltage or just something like a different thermal target default and higher speed or more aggressive fan curvature.

    Dropping HBM speeds to 800 will definitively impact overall performance as well compared to the default for Vega 64 and going into power play registry soft mods or third party utilities can be a overwhelming step too though I think Overdrive is a popular option as well leaving some of the Wattman problems or working around some tendencies like resetting to default and thinking the GPU crashed and it might be from fast boot on the OS or just some snag in the software itself.

    I've had a fairly good even if pretty cautious experience with it but I do know there were issues early on and that also applied to the Radeon VII at launch plus for Vega the overclock functionality has changed one major time with Adrenaline 2019 and 18.12.2 to where some users recommend 18.12.1 or earlier instead and then there's also been tweaks and changes to earlier drivers improving results or stability both with the pre-Adrenaline 2019 drivers and the later ones after for 19.1.1 and on.


    HBM voltage as such is a fixed value though at 1.35v for Vega 64 and 1.25v for Vega 56 I think it was and then in Wattman it's part tied to P3 from how I remember and also a sort of lowest voltage or floor value that the card drops to which is a bit complicated and explained less than ideally for what the value actually does and changes to GPU performance from lowering or increasing this.

    For HBM clocks if it's stable 1000 up from ~950 is a nice boost and then from there to 1050 and possibly 1075 or 1100 but testing for both artifacts and also thermals is a factor as dropping timing tightness if it hits the thermal limit here will impact performance and if it was to drop to 800 Mhz then that would probably seriously affect overall performance as the GPU clock is aided by the memory clock. One reason I stick to around 1500 somewhere on Vega 56 with Hynix here since the memory can't clock as well and timing tightness alone doesn't hold up performance wise so very diminishing returns after a certain point for GPU clock speed though even with Hynix hitting 900 to 950 or even a bit higher can happen so there is a bit of variance as well even if it might be different to what Samsung memory chips can do but not every Samsung will hit 1100 Mhz either or higher even if it's a popular target. :)


    There is much more for me to learn, quit a complex GPU internally and all sorts of settings and sensors and thresholds or other factors and then drivers and software side of things and Wattman and the settings for this software of the driver and controlling GPU behavior and options.

    Plus what AMD might be doing after Navi releases and focus more on these or so I expect similar to how VII got a lot of attention and has advantages in Wattman over what Vega 56 and 64 has for Adrenaline 2019 which I think range from a better zero fan and fan curvature behavior and also additional GPU hardware sensors and less P states (Three only instead of seven was it? Idle, high perf and boost then?) plus probably other stuff but I don't have the hardware.

    Probably a fun card but then cost and how it would add for a existing Vega GPU performance wise so yeah it's waiting until something newer comes out maybe seeing how little Navi here or what to call the upcoming card performs before even considering a GPU upgrade. This one is holding up nicely thus far and for my own needs the rest of the system needs a overhaul first as that's where the main bottleneck is currently. :)
     
    OnnA likes this.

Share This Page