Issue with 6900xt and undervolting

Hi guys
I'm writing here because I have a minor issue with my newly acquired asus tuf 6900xt.
Basically no matter what voltage setting i give the card either with the adrenalin software or Asus proprietary software (gpu tweak 2) it keeps bouncing up and down instead of staying at a fixed value as I've set (i would like to have 1100mV instead of the ridicolous at stock 1175).
I even made a short clip to show you guys what I'm talking about. If you watch the video you'll notice the slider being at 1100mV but the reading keeps jumping up and down (and this is not an adrenalin bug, because even gpu z shows such back and forth jumps).
I tried using either adrenalin software only while gpu tweak 2 being uninstalled, and having gpu tweak 2 only hile adreaniln was not installed (i installed the drivers via device manager).

Any of you have any idea what the issue could be? I come from a 5700xt and had none of this issue (altho that card gave me way more problems especially at launch lol) and ofc before installing the 6900xt i used ddu via safe boot to properly clean the drivers (i even checked the little "perform factory reset" box when installing the new amd drivers.
Here's the link to the clip


My pc specs are


OS
Edition Windows 10 Home
Version 20H2
Installed on ‎1/‎16/‎2021
OS build 19042.867
Experience Windows Feature Experience Pack 120.2212.551.0 (i pretty much updated it on tuesday this week, so no other updates are out yet)

CPU it's a ryzen 7 5800x stock, motherboard is a msi mag tomahawk x570 bios 7C84v15, ram are 32gb gskills trident z neo 3600 mhz cl16 ddr4 (double kit of 2*8 gb, making sure they're IDENTICAL, i used cpuz to make sure) and my psu is a corsair hxi 1000w, and it's practically brand new, i bought it for this build 3 months ago. As drives i use 2 sabrent rocket nvme and one ssd sata (crucial mx1000). Also in my original post i forgot to mention that i tried out two versions of the amd drivers, the 21.2.3 (the one i show in the youtube video i posted and the most current one) and the previous 21.2.2. Still getting the same jumps in voltages. Please feel free to let me know if i need to be more specific on anything else.

I ofc contacted asus already and they gave me the usual pr bs statement "those settings are within normal range blabla" yeah thanks i know they are. But i want to get the best performance out of my card. And having an insane 1.75 v on it doesn't help. That's almost cpu voltage we are talking about. There's no reason for a gpu to have such insane voltages, especially amd card when it's renown that their architecture works better as temperature decrease, and that doesn't help my temp decrease. Are they within normal range? Sure 70 on the core and 90 degrees on the junction are well within spec but they are definitely not optimal.
That said i watched plenty OCing vid and everyone seems just fine by moving the slider down like i did. Tho noone was using a tuf gaming. They were either reference or xfx cards.

I'm between a rock and a hard place, because i could use amazon's send back policy, i have 30 days. But it'd be sad to see they fix this issue in the next or couple next drivers releases and i gave away a great card (besides, ultimately the card still performs great, i just wish I could squeeze every bit of optimization possible).. So I'm really thorn on what to do.

Thanks in advance guys, i hope i will hear something from any of you <3

 

Seems normal to me for a 6900, mine will spike above my custom 1060v from time to time if I start going high on power limit - from reading forums wattman isn't presenting the correct mV anyways:
https://www.overclock.net/threads/o...-xt-owners-club.1774300/page-40#post-28754842
https://www.techpowerup.com/review/amd-radeon-rx-6900-xt/32.html

To be sure you'd have to modify the max voltage set by the bios with something like MorePowerTool.

With my card once I start going over 280w, the card will ramp up the voltage to compensate, probably to avoid stability issues or crashing?

Try afterburner or HWInfo instead and see.

 

Yup, i came to find it's normal behavior. What confused me, was the fact i was coming from a 5700xt and with that card undervolting was a necessity, or the jtemp would reach ridicolous 110 degrees which according to amd were "well within specs" lol. Maybe so, but im not gonna have a piece of hardware inside my case that reaches 110 degrees, even if it's just in one small point. More so if i can fix that by simply sliding a slider down and actually have better performance due to amd stuff clocking higher as temperature decrease.
So when you'd undervolt that card you'd set the voltage slider to whatever and then the voltages would not spike up and down like these new cards do, but it would stay fixed at the value you had set.

These cards on the other hand, seem to go on increments ddictated by some sort of internal algorythm. Like the voltage is seen more as a "median" value rather than a cap the card should never surpass. Like right now i have the voltage set to 1060 and in idle or light application the card does not go above 1068mV (fluctuates from 0 to 775 and tops at 1068mV). As soon as (like you said) power consumption and ofc clocks go up, the voltage seems to have some sort of increments rather than a fixed cap. Like at full power my card would fluctuate between 1068 and 1135mV, never reaching the ridicolous stock voltage set by asus at 1175mV. So yeah, im pretty much reassured now.

And no, changing software (as i said in the op i tried adrenalin, gpu tweak 2 and afterburn) doesn't change anything, which i know to be normal now since the reasons i said above. Thanks for the reply and sort of confirming what i already new, there aren't many of us owners of 6900xt around so finding data/articles is still a daunting task. Cheers man <3 take care.

 

Hello, i have a 6900 XT someone could send me a undervolt settings?

 

Hi fellow 6900XT owner. This ist normal but undocumented. I found: If you set your max frequency above a certain point, the UV-Settings will be "ignored". To proof it, try setting 2525 as a target an you'll see your voltage limit is working pretty well. This behavior sadly is the reason for some unrealistic UV/high frequency OC claims on the interwebs ^^... But most of them were reached with no practically no UV (even if set in Wattman). Enjoy you new hardware, this is a great piece of tech plus some fine wine voodoo potential. ;-)

 

there is no fine wine to be had out of navi1 or navi 2.

 

I get your point... ;-)
I think a DLSS-like implementation for Navi and some RT-optimisations could be realistically anticipated. Much more crystal ball gazing is: If CPU limiting becomes a relevant thing at some point in time AMD could be in lead...

 

AMD could be a bit clearer but instead the Wattman layout is just tweaked a bit for these cards and the community figured out how it worked after some trial and error and sites like Igors Lab doing some testing on it, testing so far makes it so the GPU's operate on a sort of voltage offset mode but using Wattman you have to fine tune the clock speeds as well if you want it to actually operate around the lowered value instead of potentially boosting up to default/maximum which it can otherwise still do.

More Power Tool and defining a lower default wattage and voltage level works though testing will be required to ensure it is stable and under additionally demanding conditions like ray tracing being in effect or with AMD SAM / ReBAR active.
My own testing was basically to use the 6800 in a game see what it scaled up to and then what it set for the power draw.

Lowering the clock speed until the power draw and voltage fell below the maximum and then from there finding a stable offset value without getting the GPU to scale it back to maximum speeds.
Doing it like this though has very limited headroom without lowering clock speeds down a bit, would imagine the additional hardware on the 6900XT also demands a bit more whereas the 6800 could scale but has lower caps from AMD's side instead.


Assassin's Creed Valhalla makes the GPU scale well and ReBAR, SAM as AMD calls it puts some nice demand on the GPU making for a decent stress test.
Without fine tuning it via More Power Tool there wasn't much to be done from the default settings without dropping down the GPU clock speeds a fair bit it peaks pretty quickly to the maximum and scales a little bit further with additional power draw allowed.

-25 offset and around a 1.0v was about what I could get for avoiding driver crashes when the GPU tried to peak and momentarily hit higher speeds where it needed voltage to remain stable, 6900XT has a bigger ceiling though so chances are it can be tuned at least somewhat better for it's 1.175v cap and a few higher bios settigns over the 6800XT for power draw and such though having the full 80 core units perhaps offsets it somewhat


Or using More Power Tool from a default/blank Wattman profile and testing if the enforced modded limits are stable which will allow a lower setting though then you have to remember that this is your new default that any Wattman changes will operate from instead so you don't hit any instability here.


EDIT: It can overclock a bit but undervolting takes some fiddling as a result of how this offset mode operates and since you can only lower the settings and not push them higher or define a better curve for it.

And how the GPU can still hit the maximum anyway if it's required so it takes some tweaking unless you utilize third party software and set the limits that way instead.



EDIT: The system really could do with more advanced settings if it's intended as a complete overclocking software suite.

Curve editor and setting minimum and maximum levels maybe a mid preferred value along with toggles for if you want to adjust the base value or operate on a offset mode as well allowing users to fine tune the GPU voltage and it's power draw for example.
The various locks on adjusting the power draw both up and down and the maximum allowed before the GPU goes into some 500 Mhz safe mode or minimum mode really could be done away with too but I don't see that happening.

 

I run 1100mv in my favorite 24/7 profile. 300w MPT & +15% in Wattman. Clock dialed in for ~2500MHz permanent real speed in Timespys GT2 loop. Fan curve dialed in for 40 dba noise in GT2 - translates to 69°C GPU edge temp. This earns about 20500 graphics score in Timespy on a "non SAM" system. Memory: 2126MHz fast timings. If i set mem to 2150MHz I lose 500 points - most likely due to "invisible" errors. This profile is (multi-)gamestable since days.

I'll spend some more time OCing in the future. Hope to see more community results soon. Maybe at some point in time BIOS mods or tools like MPT can lift the memory clock limit (and more).

 

I undervolt like this: ~1150v setting, but limit clockspeed from to 1990 to 2090. With my GPU I can archive a great undervolt this way - with 875mv to 900mv real load voltage. It stays well unter 200w in most cases and clocks at ~2040mhz (mem 2126 MHz fast). This setting is able to archive a solid 17500 graphics score in Timespy. A pretty satisfying efficency in my eyes.With some curve tuning the noise stays below 36dba @ 65C Core / 78° Hotspot temp. It's fiddly to create nice profiles - but I really like the results.

 

can anybody please tell me the idle temperature of 6900 xt, especially in tuf 6900 xt, in a full tower case while only cpu fan working like during web browsing? i dont use any airducts inside the tower. if you undervolt this card or limit max frequency or limit fps, will the idle temperature decrease? if yes by how much? answers would be appriciated, thanks

 

Hi
with auto undervolting (1150V) with the fan at 820rpm the edge tem is 34 and junction at 36, at stock with the fan at 0rpm same edge temp and junction at 37.
Undervolting at idle does nothing because the voltage is automatically very very low, the GPU is almost completely off.