1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

RX Vega Owners Thread, Tests, Mods, BIOS & Tweaks ! (cont.)

Discussion in 'Videocards - AMD Radeon' started by Glidefan, Mar 31, 2019.

  1. Daeteria

    Daeteria New Member

    Messages:
    9
    Likes Received:
    5
    GPU:
    V64 Nitro+ LC
    Hi again! After following the thread now and again, I got intrigued about my hotspot temps as well. As it turns out, my waterblock mounting on the Nitro+ wasn't that perfect the first time out:

    [​IMG]
    After a Superposition 1080p Extreme run, max temps were:
    GPU: 47 °C
    Mem: 44 °C
    GPU hotspot: 82 °C

    So I did a repaste and remount with somewhat more pressure as well. Results with identical settings:
    [​IMG]
    Now the max temps were:
    GPU: 46 °C
    Mem: 43 °C
    GPU hotspot: 72 °C

    So roughly a 10 °C difference with the hotspot peak temps, I'd say quite a difference! However, a 26 °C delta between the edge and hotspot temps still seems quite steep. For reference, I also did a run with 1666MHz / 1055mV core (my usual go-to setup) and ended up with 43 °C edge / 41 °C mem / 66 °C hotspot, so a 23 °C delta.

    Can any of you guys with Vegas under water post your edge-hotspot deltas? After researching a bit online, I concluded that my results are now within reason, but it'd be nice to have some additional validation. :)

    Cheers, Daeteria
     
  2. The_Amazing_X

    The_Amazing_X Master Guru

    Messages:
    332
    Likes Received:
    112
    GPU:
    Red Devil V64
    That's so high on the hot spot. I have 60c core and 80c on hot spot
     
  3. DukeNuk

    DukeNuk Member

    Messages:
    16
    Likes Received:
    4
    GPU:
    Vega 56 Nitro+ LE
    those temps on water block are definetly not right. i have 66 core and 85 hot spot on not modified Nitro+ air cooler, with core at 1810mhz. in my case, fans are not going higher than 2000rpm (about 60%).

    other thing is you power consumption. it's really high imo. 320w at 1700mhz, that's a lot. my card takes 264w.
     
    Last edited: Aug 23, 2019
  4. The_Amazing_X

    The_Amazing_X Master Guru

    Messages:
    332
    Likes Received:
    112
    GPU:
    Red Devil V64
    power wise its the same as mine. to get even close to 1600mhz I need 360w. maybe we got crap dies.
     

  5. DukeNuk

    DukeNuk Member

    Messages:
    16
    Likes Received:
    4
    GPU:
    Vega 56 Nitro+ LE
    you missunderstand. my card take 264w with core set to 1810mhz. that's a lot lower than yours. more power=more heat. maybe that's why your hot spot is so high.
     
  6. Daeteria

    Daeteria New Member

    Messages:
    9
    Likes Received:
    5
    GPU:
    V64 Nitro+ LC
    I think you're power throttling the card if you only hit 264 watts at those clocks. Of course I could set the power limit to whatever and set the target clock for P7 to whatever else and theoretically have "1700MHz at 100 watts", but in practice the card would never achieve these clocks. You can only verify your actual clocks by monitoring them.

    For reference, I plotted some data a couple of weeks ago indicating clock speed / power draw relationship:
    [​IMG]


    Another perspective on the sama dataset:[​IMG]

    As a sidenote, that 1810MHz (set) clock is pretty intensive. My card maxes out at a set value of ~1745MHz @1200mV, any higher clocks and it will crash. The actual clocks vary depending on the load of course, but max out somewhere in the 1720s maybe.
     
  7. DukeNuk

    DukeNuk Member

    Messages:
    16
    Likes Received:
    4
    GPU:
    Vega 56 Nitro+ LE
    i do monitor them during test. under full load clocks are between 1723-1836 to be precise. done some testing with higher power limits, consumption was going up, yet clocks were same.
    for another expamle, my everyday settings are at 1632 P7 with ~190w power consumption, clocks are nog going lower than ~1550.

    i will do more testing to be sure i haven't miss something. there is always possibility of wrong readings as i have some driver issues, which i mentioned few posts earlier.
     
  8. Daeteria

    Daeteria New Member

    Messages:
    9
    Likes Received:
    5
    GPU:
    V64 Nitro+ LC
    Alright. I'd still stand behind the clause that under very heavy loads such as the Superposition benchmark, the clocks will throttle if not given enough power.

    Out of interest I tried out a lower power limit and ran a lighter workload, in this case Diablo 3 at 1440p / max settings / 142 fps framelimit for about 2,5 hours. I didn't log this session to file, so don't have any graphs to show. I did monitor the clocks through wattman's overlay, and yes, it actually boosted to over 1650MHz with the power draw in the low 200s. However, the gpu load levels never went over 90% either, and usually hovered in the 60 - 80% region, so no throttling could occur.

    Here are some hwinfo stats:
    [​IMG]
    The GPU Clock max value is bogus as usual, but the interesting numbers to look at are the temps and power draw. It is quite evident that the GPU hotspot temp has more correlation to the power draw directly than to the GPU edge temp. In this case, a max power draw of 246 watts resulted in only an 18 °C delta between the edge and hotspot temps, which is more in line with your results DukeNuk.

    I'll try to do some logging tomorrow with Superposition, running different power limits but otherwise same settings and see what the differences turn out to be.
     
  9. The_Amazing_X

    The_Amazing_X Master Guru

    Messages:
    332
    Likes Received:
    112
    GPU:
    Red Devil V64
    Im thinking of puting my card under water,

    What do you guys think of this block.

    https://www.aquatuning.co.uk/water-...amd-rx-vega-m07-incl.-backplate-black?c=22352

    Edit: So I think i did apply the powertable, MorePowerVega64devil

    Run DDU and then apply the PP and then instaled drivers.

    Still waiting for drivers to finish

    Edit 2: Nope still cant Apply the PP

    Edit 3:Got it to 150% lets see

    Ok got a peak of 1713mhz, after some seconds the driver crashed, I think the hot spot is just to hot, card was runing at 70c and hot spot was at 106c, to hot.

    Still the 1600mhz was broken to a 1620mhz lol. 366w Amd drivers
     
    Last edited: Aug 24, 2019
  10. DukeNuk

    DukeNuk Member

    Messages:
    16
    Likes Received:
    4
    GPU:
    Vega 56 Nitro+ LE
    right, after many houers (finished about 2am) i found some issues and (i think) managed to fix all of them. long story short, corrupted motherboeard bios, in a really,really weird way. took a lot of reflashing, ended up with few months old modded version. seems to work fine (no BSOD with new drivers).

    it was affecting my reading, well partly. HWinfo was giving mentioned clocks reading yet card wasn't working on this clocks but pwer draw wasn't very far from correct. i have no idea how mb bios could cause it but it did.

    any way, done UV over again. didn't manage to reach ~1800mhz (~1760mhz max) yet it's hot day so maybe other day. card was at 69°C core and 89°C hot spot with clocks averaging at 1700mhz with 289w power draw. even maximum clocks went down, rest is working with lower power draw. my low gaming plan averaging at ~1500mhz is now at ~130w. to be sure there are no miss reads i've checked readings with msi afterburn osd overlay. ofc it's all valid only if readings are correct this time, which i hope for.

    i'm very sorry if i coused any cofusion, i didn't expected such weird mb issues.
     

  11. apple020997

    apple020997 Active Member

    Messages:
    54
    Likes Received:
    24
    GPU:
    RX Vega 64+waterbl.
    Just FYI, my Vega 64 (EK-FC Radeon Vega non-RGB) at 100% stock settings in my full custom loop with 2x240mm rads (Alphacool NexXxos ST30) never reaches over 55ºC hotspot and 50º core with fans (4x EK Vardar Evo RGB)/pump (Jingway DP1200 PWM) at around half speed (they’re controlled by the CPU temp), and with my maximum stable OC never over 55 core and 60 hotspot, with fans and pump at full speed. BTW, my CPU block is extremely restrictive (Eisblock XPX) and half of the front rad is partially blocked by the HDD cage/PSU shroud (my case is an Aerocool Aero-800)...
    Be very careful to apply correctly the TIM on the GPU/HBM dies, or hotspot/HBM temps will skyrocket, air cooled or water cooled. It’s an enormous die, so a cross on it and a cross on each HBM die is good (EK instructions)
     
    Daeteria likes this.
  12. Daeteria

    Daeteria New Member

    Messages:
    9
    Likes Received:
    5
    GPU:
    V64 Nitro+ LC
    That is pretty weird indeed.

    Ok, so after a couple of hours of testing and data formatting, here are some line charts as promised going from highest power settings to lowest. I used the same axis scaling in each chart for brevity.

    [​IMG]
    Pretty stable clocks as expected. The temp delta is in line with my previous results.

    [​IMG]
    At +10% power limit we can already notice huge jumps in the clocks. The temps seem to scale quite linearly here.

    [​IMG]
    At -10% power limit the clock frequency zigzag seems to trend towards the lows rather than the highs like previously.
    Here we can see a pretty dramatic dip in the temp delta; while the edge temp only dropped by approx. 3 degrees from the previous settings, the hotspot temp went down by 9.

    [​IMG]
    As a little bonus, I also ran once at 1616MHz / 955mV / -20% p-limit with fans (1800rpm) and pump maxed out.
    Here the clocks are quite stable again, since the power limit is more in line with the clock target and voltage. The temp deltas have narrowed once again and pretty linearly with the power draw as well.

    I think these results somewhat verify my previous assumptions. A higher power limit enables a more stable core clock (common sense really and verified elsewhere on multiple occasions), but also increases the edge to hotspot temperature difference significantly.

    How much this actually matters is another matter entirely, but I'd dare speculate that keeping the hotspot temps under the throttling limit is enough performance-wise. For long-term durability, lower temps are usually considered better. The power draw to performance ratio (as in extra fps) has huge diminishing returns, i.e. keeping the clocks stable with 300 watts doesn't really yield that much more performance than the fluctuating clocks provide at say 250 watts. However I'd say the more stable clocks could give better frametime stability overall (less stuttering and fps fluctuations), especially against an fps limit.

    All in all, an interesting-ish experiment. Hope someone finds this useful!
     
    Exodite likes this.
  13. The_Amazing_X

    The_Amazing_X Master Guru

    Messages:
    332
    Likes Received:
    112
    GPU:
    Red Devil V64
    I really don't know what's going on with mine -5 power and 1030v does 1570mhz stable no issues 220w max sometimes 180w, from here on things get stupid the more I push the card the less I get and only if I push to stupid power limit 150% do I get over 1600mhz. Will probably order alphacool Vega M07 block.
     
  14. DukeNuk

    DukeNuk Member

    Messages:
    16
    Likes Received:
    4
    GPU:
    Vega 56 Nitro+ LE
    when doing my setup again, i've notice that clocks were not doing well with too low V but as well with too high. in my case to keep average target clocks stable i was playing with p5,p6 and p7 for long time to get right V, to get stable clocks.
    [​IMG]
    for expamle, that's my low game plan. even there is difference of 70mhz between P5 and P7 clocks are staying at ~1500-1520mhz all the time. i've decided to target stable average than jumping clocks. is it good decision? i don't know, maybe some1 more experienced will judge.
     
  15. OnnA

    OnnA Ancient Guru

    Messages:
    9,493
    Likes Received:
    1,941
    GPU:
    Vega 64 XTX LiQuiD
    On -20% (or similar up to -15%) You'll have lower clocks and not so good performance.
    Best is to max -POW at -5% or -8% -> not too much hit on performance.

    Test it at 0%, +8% and then -5% and -8% and You'll see.

    Up to -25% is good for Older Games or Card games like MtG Arena etc.
    Up to -5% is good even for BFV Multi (you got W at max 180tW when Play it's max not average, average is ~70-80tW ;) )

    [​IMG]
     

  16. Exodite

    Exodite Ancient Guru

    Messages:
    1,911
    Likes Received:
    164
    GPU:
    Sapphire Vega 56
    All Vega silicon is different, obviously, but there's a correlation between power, voltage and frequency that you kinda have to experiment your way towards finding the ideal spot for your chip.

    Unless you're power limited, ie. negative power limit and hitting it consistently, your numbers seem to indicate that your card isn't getting enough voltage to hit higher frequency. Try increasing p6 and p7 voltage very slightly, say 1010 and 1030 mV, and see what the same benchmarks/games give you in the way of frequency.

    In my experience the stepping/delta of the power states also affect how aggressively the card will try and ramp voltage to hit boost clocks. Ie. of you have 3 mV steps it won't ever try and push voltages very far while the default ~50 mV steps will quickly ramp the card past its power limit.
     
  17. Daeteria

    Daeteria New Member

    Messages:
    9
    Likes Received:
    5
    GPU:
    V64 Nitro+ LC
    I wouldn't pull the trigger on a waterblock just yet and try to figure out what the exact problem is and what it affects if anything.

    So the clocks you're talking about are monitored clock readings, not set values? If that's the case, what are you monitoring them with? In my experience, Wattman and HWMonitor report similar values to each other. I'd be tempted to say that these are closest to the "true" values. GPU-Z and HWiNFO64 seem to report similarly to each other as well, both reporting higher clocks than Wattman or HWMonitor and always reporting higher peak clocks than my set values, which doesn't seem right.

    As stated many times over, every Vega chip is different, so direct clock/voltage comparisons cannot be made. Anyways, 3dmark scores provide a pretty good validation method and if you match the results of other people, your card is doing just fine.

    For a point of reference, here are some of my Time Spy graphics scores:
    1616MHz / 955mV / +50% p-limit
    - 7735 graphics score / GT1 53.77 fps / GT2 42.05 fps
    - Reported readings (HWMonitor / HWiNFO64): 1542MHz / 1584MHz / 182W


    1666MHz / 1055mV / +50% p-limit
    - 8158 graphics score / GT1 56.73 fps / GT2 44.33 fps
    - Reported readings (HWMonitor / HWiNFO64): 1645MHz / 1710MHz / 240W


    1700MHz / 1128mV / +50% p-limit
    - 8339 graphics score / GT1 58.02 fps / GT2 45.30 fps
    - Reported readings (HWMonitor / HWiNFO64): 1693MHz / 1739MHz / 290W


    Note that all runs were done with the same memory settings that can be seen in a previous post (1100MHz, custom timings).

    Another point of reference for default settings, albeit with older drivers:
    https://www.guru3d.com/articles_pages/asus_radeon_rog_rx_vega_64_strix_8gb_review,30.html

    I'd say try this one out with DDU'd fresh driver install and no pp-table messing about first. Then go for undervolting at default clocks and see where you end up.

    Cheers, Dae
     
  18. DukeNuk

    DukeNuk Member

    Messages:
    16
    Likes Received:
    4
    GPU:
    Vega 56 Nitro+ LE
    that's exactly what i said. i want it to stay stable at 1500mhz without "jumping" around. i can inrease voltage for higher clocks obviously, which i do for my higher plans. i just decided it will be better to have max clocks staying stable. i have ~1600mhz and ~1700mhz plans which are set to stay on this clocks average under load. ofc minimum and maximum clocks are different and changing all the time, i want to have average stable around my target value. again, is it good ide? i don't know :)
     
  19. OnnA

    OnnA Ancient Guru

    Messages:
    9,493
    Likes Received:
    1,941
    GPU:
    Vega 64 XTX LiQuiD
    ^^ You can't make Polaris/Vega/Navi to not jump clocks.
    That's the way ATI GPU works, when needed it rises the P-State.
    We have 7-Pstates for Vega, i've noticed that in TS it jumps from P-6 to P-7 but mostly stays on P-7
    The same behaviour can be observed in Gaming, especially with FRT or Chill.
    That's the way it rocks.

    UPD.
    It is always ~ +/- 15-30MHz less than target P-7
    I have P-7 at 1668MHz 1.093v and it stays at ~1640MHz 1.087v (when i play BFV)
    It's best for Perf vs tW for my Unit.
     
    Last edited: Aug 25, 2019
  20. DukeNuk

    DukeNuk Member

    Messages:
    16
    Likes Received:
    4
    GPU:
    Vega 56 Nitro+ LE
    i know. i just set it to not jump too much. for example average clock stays around ~1500mhz but p7 is at 1572mhz. that's what i mean. i dont' want it to stay at same clock, i just want average clock to at set target.
     

Share This Page