Discussion in 'Videocards - AMD Radeon' started by OnnA, Oct 29, 2020.
Well, it's like mine card, there are worse performers but also better ones as we know.
Was going to put it on a waterblock. But with these numbers and the low power bios im not going to do it.
If at some point we can put a high power 6900xt bios on it maybe.
There are 2 ways to do undervolt.
First limit your max gfx voltage by more power tool. (Which will eventually cause to crash in certain scenarios)
Second; Create a bug with wattman. Like here:
As you can see when the slider at the lowest point gpu voltage fluctuate 1.0 to 1.125. If i set more voltage by slider it starts to fluctuate btw 1.025v to 1.150v. The best possible performance only can be get with lower voltages.
Also wattman is not working as is intended. Overclocking with Amd card require master level of understanding what's really going on.
Nvidia is really simple. Even 10 years old kid can tune it nicely. That's why i will chose an nvidia card next time. Amd cards takes lot of time to tune.
Using Sam require much more voltage. Sam off&950mv 6800xt can get 2400-2500 and "maybe" 2600mhz in my opinion. But didn't tried
Since the frequency not really determine performance and important for Amd cards no need to try higher.
Edit: Max gfx voltage 950mv by MTP here:
Above that frequency it's crashing after 2 or 3 minutes. So 2600mhz is not possible
Ahhh so that's another reason. I have SAM enabled with my 5950x and 6900 XT. Good to know.
^^ So now we know that RDNA2 can sustain 2500MHz at only 950mV
That's a very good power envelope, runs @ around 230tW.... 1440p King for sure.
Plus i tried lowest possible voltage which is 881mv and here is the result:
This provides %10 lower than at Sotr @2750mhz
2300MHz at ~190tW
Mine does 2600mhz at 975mv, more clocks dosent crash but I start getting artifacts,
So I think thats the limit of mine
Maybe with a diferent bios.
Still trying the lower limits of gpu just for fun
800mv+boost to 2070mhz+Sam enabled 4k Sotr Highest Settings+ Gpu power 150w
Well I think I will need a 1000w psu. My 750w gold psu turns off if I push the 6800xt to the max on firestrike 4k . Maybe its allot for the 3900x and 6800xt to triger the psu.
With the Vega it was fine even at 380w
That seems odd unless it's older and is limited in efficiency the power draw is around 300 - 350W for heavier activities pushing the CPU a bit high you'd still have 300W or so at least for the PSU's total capacity.
(TDP of 250W for the GPU core and then the rest so 300W roughly and then a bit extra or if overclocking and using the extra 15% which can push it a bit higher still.)
750W Gold effectiveness that's 80% Plus I think?
It changes though there's a new Titanium above Platinum so 90% and 95% I believe but it can vary a bit.
3900X can hit upwards of 150W depending on how you're pushing it with PBO or not and then say 350W for the GPU when pushed hard.
That's still 200W and a bit extra unless it's something in regards to the Amps and tripping up the PSU but the Vega GPU was really hard on that so if the PSU could handle that card this seems a little strange.
(Vega 64 is close to 400W if not more with the total and not just the GPU itself.)
EDIT: I think SAM kicks it up a bit more though come to think of it, not just about needing more if you've undervolted the card it also pushes it a bit harder by default.
EDIT: Also I need to double check so I'm not mixing the 6800 and 6800XT here and then various measurements and results and testing.
Think I got it right though it's about 100W difference between the two so 200 up to 250 -> 300 up to 350 and then a bit higher during peak loads so up to of 400 bit more depending on if it's overclocked giving it a bit more power draw or not.
AMD kinda limited it by default but I suppose an easy test is to pull the power draw to -5% or so depending on where it's limited at.
8% I think for whatever reason and then up it's 15% but why limit it so low when lowering power draw the same 15% would be good if not 20% since you're not exceeding the maximum possible GPU power draw here.
Also need to re-check how this is monitored part of it's going to be AMD's GPU readout others might be using instruments here for the actual full value.
(TDP +50W kinda roughly put I think is how that went.)
Sigh well the system shouldn't be pushing over 500W anyways so in that regard the PSU should be fine.
I tend to go fairly high-end on the PSU myself though but limited availability here and long as it's not 1500W and the big price increase for these extreme models it's usually OK even though power draw wise the 1000 - 1200W models are far more than enough also giving a few years of slowly losing peak effectiveness but then it's time to replace it anyway best not to risk anything though I am often pushing 6 - 8 years which is fairly solid, so far at least.
("Well it's been five years better blow a circuit now." It's going to happen at least once just the way it has to be these universal laws and all.)
Guess there's More Power Tool as well but for testing you'd leave the Power Draw slider at 0% to begin with instead of adding a extra percentage or the full 15% to a new untested base value and peak maximum draw this can lead to and from that and seeing how the GPU scales and what it draws without anything extra, probably a little different though exceeding the maximum the GPU can draw and having it shut down from that though I suppose depending on what you modify if it's just the TDP GPU Power Draw Total or if you also alter other settings like the ampere and current settings you might hit a PSU protection trigger or GPU safety shutdown if it hits some spike or peak load above what it can sustain.
Haven't really done anything with that yet just something I came to think about because there's a lot of info about it but little in the way of testing and monitoring other than some of the later Igor Lab articles and then user posts and info on their results and experiences like some of these posts here.
EDIT: Course Vega 56 and 64 both kinda hit high peak loads for wattage and ampere so if that was just ticking along the PSU should be capable of handling the 6800XT too.
(XT's a bit closer to Vega but the peaks are more subdued not 400W+ levels of short bursts of power. )
That's the Water edition though but yeah.
Update on the tweak/tuning/flashing tools from hellm (one of the lead devs of R.B.E.) :
rough translation from german that maybe is not exact but contains the original message in an unbiased way :
We have just flashed the first BIOS-Mods on the 6800XT. Unfortunately without success. We were able to find a flashing tool that was Linux-based and actually worked with Navi21, however after the flashing we were unable to make the card boot. On top of that not only there was no display output at all but literally the card stopped to communicate with the system and the only way was to flash it with an external programmer.
So instead of good news I want to share a warning with you. We will still work on it, however before we will dare another try we need to learn more about the Navi2x firmware part.
Many thanks to all the ppl that have helped, especially L!ME who was not only chasing after a working flashing utility but also brought an external programmer. Thanks of course also to to dman ersa04 Tymerias and all the other ppl that have upped their BIOSes here.
By the way --- the Navi21 BIOS is 832kB in size, it seems that GPU-Z is reading/extracting only the first 512kB.
Update an alle, wir haben soeben die ersten BIOS-Mods auf eine 6800XT geflasht. Leider ohne Erfolg. Wir hatten eine Linux-Version gefunden die tatsächlich auch mit Navi21 funzt, aber wir konnten die Karte nicht mehr zum Booten überrreden. Und dabei wurde nicht nur die Bildausgabe verweigert, die Radeon wollte überhaupt nicht mehr mit uns sprechen und konnte nur noch mit einem externen Programmer geflasht werden.
Also, statt guten Nachrichten habe ich nun eine Warnung auszusprechen. Wir werden weiter daran arbeiten, aber bevor wir uns wieder an Navi2X heranwagen müssen wir noch etwas mehr über diesen Teil der Firmware in Erfahrung bringen.
Unser Dank gilt allen, die hier bereitwillig ihre Hardware zur Verfügung gestellt haben. Im besonderen @L!ME, der nicht nur die Suche nach der passenden Software entscheidend vorangebracht hat, sondern auch einen externen Programmer einbringen konnte. Und natürlich @dman @ersa04 @CS9K @Tymerias und allen anderen, die hier ihr BIOS hochgeladen haben.
Übrigens ist ein Navi21 BIOS 832kB groß, GPU-Z scheint nur die ersten 512kB auszulesen.
Thoughts (including personal bias/opinion):
-Currently extracted BIOS files might be trash in case some crucial model-dependent information (like memory, timings, crucial firmware parts etc. etc.) are stored in the remaining 320kB, and we would probably need to wait not only for a flashing tool, but also an update on the reading/extraction tool.
-We will definitely see it as there are more and more ppl involved and eventually there will be a leak of the lower-level tools that could be debugged.
-Currently safe way to tweak the cards is with the use of software methods -- MPT and Wattman (eventually Afterburner etc.). There is currently no possibility to circumvent the limitations set in the firmware (e.g. max clocks etc.).
Is it bad ? Well... not really... at least not for the majority -- 99.9% of owners....
For them it is enough to OC the hell out of the cards and reach the power delivery limits of ca. 370-375 Watts. It can be easily done with a few clicks in Wattman, and in MPT.
Only die-hards with 3x 8pin / modded power deliveries AND custom loops would want to push the limits even further... or rather --- to find the limits of Navi21 cause they weren't found ...yet.
Will keep you posted once there are some new infos.
it seems like ive got a big dud of a 6800 XT highest core clocks i can get with default voltages (anywhere with the voltage slider, mpt 350W) is 2410mhz - 2610mhz core with standard nitro+ cooler. anything higher and its pretty much an instant crash in timespy (https://www.3dmark.com/spy/17339202) just letting you guys know how massively variable the silicon can be.
MPT 350w --- are we talking about the "Power Limit GPU (W)" value that you set to 350W ?
Then definitely this might be the problem as you just set the cards total board power maximum at ca 400W.
Cyberpunk 1440p ultra preset 6800xt undervolt Note that due to video capture consumes 10w higher& and few fps lost
watching hwinfo64 with timespy running at most i see 350w for gpu PPT/asic power, with hot spot only hitting a peak of 102c on a hot day. previously had my vega 64 running a 142% powerplay table on water. planning on putting a waterblock on this card too.
Lol that seams like my 6800xt gaming x trio.
This cards need more voltage.
The Power Limit GPU in MPT is the GPU _only_ part.
Add to it ca 45 Watts for GDDR6 and VRMs and other losses. Basically instability might be because of the power delivery problems.
2x 8 pin + PCIe spec is maxing out at 375 Watts. By setting the GPU alone to 350 you are hovering around 395-400 Watts... That is an OVERKILL.
And I am not even starting to think about the possibility that it might be possible to add additional (up to) 15% in Wattman.... (455-460 Watts with the slider to the right... lol)....
Either you really have bad ASIC quality OR --- incorrect assumption that this is the actual total board power.
I would suggest before coming to conclusions to lower it to 280W in MPT, reboot and retest.