AMD RED BIOS EDITOR und MorePowerTool - BIOS-Einträge anpassen, optimieren und noch stabiler übertakten | Navi unlimited
- Themenstarter Igor Wallossek
- Beginndatum
Veii
Veteran
- Mitglied seit
- Okt 24, 2021
- Beiträge
- 184
- Bewertungspunkte
- 212
- Punkte
- 43
Sind leer
Deswegen kannst du sie nicht nützen // mit nem anderen Bios oder ner anderen MPT, geht es
Potentiell könnte ich was machen
Aber eigentlich warte ich auf Hellm's MPT update
ShirKhan
Veteran
- Mitglied seit
- Jul 23, 2021
- Beiträge
- 203
- Bewertungspunkte
- 140
- Punkte
- 43
Danke, nicht wegen mir.Potentiell könnte ich was machen
D4NNY_143
Veteran
tut mir leid, aber ich habe das jetzt 5 mal durchgelesen, sorgfältig und mit dem wissen welchen ich noch von Polaris, RDNA1 und meiner aktuellen karte habe, aber habe rein gar nichts verstanden, ich gehe mal davon aus, dass die Option GFX_per_part_vmin eine Art automatisches UV ist?
Dass die Karte wie Stock sich für jede Mhz eine eigene Spannung ziehen kann will ich ja behalten, dann ist das nicht wie Polaris mit den States.
Ich weiß halt nicht mal ob die GFX sich im idle ausschaltet und den SoC alles laufen lässt oder ob da immer Spannung an steht, sonst würde ich ja mal versuchen beim niedrigsten Takt schon zu UV, das hat bei meiner 5700XT mit 650MHz gut geklappt.
Letzten Endes will ich nur erreichen dass die Karte bei dem OC Profil also nem Takt von 2700MHz die mit 1.15V laufen, umgehen kann, aber mit Standard Setting bei 2200Mhz läuft wie momentan, allerdings stehen keine 900mV an sondern 840mV um den dreh, was mir sehr wenig erscheint und spiele wie warzone starten nicht, aber die meisten anderen wie Fallguys, WoT, etc laufen über stunden prima.
Bis sich auch die spiele ohne Fehlermeldung schließen.
Den Takt habe ich auf der selben Spannung um 300Mhz erhöht, und die Spannung auch um 90mV bei 2200Mhz runtergesetzt, und es lief durch alle Benchmarks, plus 9 Std Valley Benchmark.
Ja, da schraube ich auch noch dran rum, die Meldung kommt viel zu oft. Ich schreib dich vielleicht mal an wegen ner Wiederholung, sollte ich aber selbst hinbekommen. Außerdem erstmal MPT und PerPart, werde ich bis zum Ende der Woche schaffen.
Veii
Veteran
- Mitglied seit
- Okt 24, 2021
- Beiträge
- 184
- Bewertungspunkte
- 212
- Punkte
- 43
Geb mir genaue instructions. Was wo geändert wurde.
Ich denke der Fehler liege im Verständniss der Slider bzw Werte.
881mV VID , sind genug für 2019Mhz Strap
Es müssten nahe zu 931mV anliegen für 2219 Strap.
Etwa +25mV alle +100Mhz, so um den dreh herum. // +/- 6.25mV je nach binning
Wären nahe 1056mV für 2719 stock.
(welches mir XTXH und KXTX Leute auch bestätigen können)
In deinem Post habe ich keinerlei Navi2x erwähnung gefunden, noch nen ASIC %
Ich kann dir leider nicht Folgen was du wo auf welcher Karte, falsch machst.
Sind das VID, sind das SVI2.0 Werte durch HWInfo. Ich kann aus deinem Post leider zu wenig entnehmen.
EDIT ~ wäre für dich womöglich Sinnvoll
Zuletzt bearbeitet
:
Voltage min/max in MPT "power" tab.
Clocks are in MCT.
The voltage slider in wattman (or Voltage in MCT) would change the curvature degree. Use these together with PPT limiter.
Actually, you don't really need to touch anything besides mentioned above cause the GPU power management is very smart nowadays.
Usually tau means "Time Constant", that (i think) is the response time needed for the load transition to settle in ac-dc or dc-dc converters.
With regards to the GPU, I beleive the meaning is the same as for the "Slow PPT Time Constant" used together with the "Slow PPT Limit" in a Ryzen Mobile parts, taking into account the power source (ac or dc). Its well described in amd's own slides for the STAPM tech.
Why so? "Per part" just mean some values are set individually per ASIC. The same way as fuses programmed at the ATE stage.
Zuletzt bearbeitet
:
Veii
Veteran
- Mitglied seit
- Okt 24, 2021
- Beiträge
- 184
- Bewertungspunkte
- 212
- Punkte
- 43
Ahm,
PW-Linear followed LkgID
Per Part is mostly Navi1x Quadratic Curvature but more granular in Voltage ranges.
Unless you mean ASIC (SKU) ?
Outside of PP-PW-Quadratic, SKU limits apply ontop of per Bios PSM delta (fused in PSP or generated elsehow, unsure but has to do with ChipID)
I think it's used because it's more granular and better for clock gating.
I got both PW-Linear and PP-PW to nearly the same values, before checking how far of KXTX SKU is in delta @ same LkgID.
Downside of full PW-Linear is the curvature shape and the reaction time. Also the lower voltage stability
On the highs, let's say post 1050mV it's "ok" very similar
But on the efficiency side, it's worse & faster to be unstable, although easier to work with (with fixed freq voltage droop margins)
Vs that but times 5 as smoother curvature.
I could be wrong, but so far i haven't seen LkgID bothering much if not at all & so presets are WIP between SKUs
Especially since there is quite some thermal and voltage headroom left from AMD.
But i could be wrong, only that "data" shows subtle differences. Subtle significant.
@PJVol are you busy till the end of this week ?
I'd need your time
D4NNY_143
Veteran
Hier mal die MPT datei, wie meine RX6600XT läuft.
Das ist der springende Punkt, es liegen nur 0.85V an.
Die Karte ist jetzt so wie sie soll.
Leise bei 1300 RPM im Volllast, Kühl mit 75°C Hotspot und Sparsam bei max 90W ohne leistung zu verlieren.
Das ist der springende Punkt, es liegen nur 0.85V an.
Die Karte ist jetzt so wie sie soll.
Leise bei 1300 RPM im Volllast, Kühl mit 75°C Hotspot und Sparsam bei max 90W ohne leistung zu verlieren.
Anhänge
Well, as I see it...
All these "piece-wise" or "quadratic" or whatever are just math models used for calculating coefficients via linear regression, that needed (if at all) to program transfer function for the "dumb" PSMs to make their output Vdd-to-Count characteristic linear, and I beleive these calculations are made during BTC, not in real-time. I think you might confuse piece-wise VF points with a DPM P-states, as it was in Polaris. Navi doesn't use discrete states, aside from two, defining Min and Max curve points. PerPart model (3) is also piecewise linear (according to linux sources), and I see the difference is in VF ref. points are available from PP Table. More confusing that they're not, with the default model 1 (piesewise linear pp)
#define NUM_PIECE_WISE_LINEAR_DROOP_MODEL_VF_POINTS 5
typedef enum {
PIECEWISE_LINEAR_FUSED_MODEL = 0,
PIECEWISE_LINEAR_PP_MODEL,
QUADRATIC_PP_MODEL,
PERPART_PIECEWISE_LINEAR_PP_MODEL,
} DfllDroopModelSelect_e;
Actually, more recent PSM implementations (based on current controlled oscillators or built-in reference oscillator as clock source) are of low latency and insensitive to temperature variations, so no need to store calibration data into fuses or nvram. Though judging from the Sienna Cichlid PP Table, the actual PSMs seem not to be as advanced )), as those presumably designed for the gfx11.
Idk, still have some work to do, we'll see as it get closer
@hellm
Do you think it's still a good idea for that "Dfll model select"?
With the 6800/6900 SKUs the driver stopped functioning when 3 is selected.
Regarding MCT (freesync, frtc and other stuff), I currently using minimal adrenalin setup, which has all features I need except SAM. In BIOS it is enabled (4g pages), but its actual runtime status in OS is nowhere to be seen.
Zuletzt bearbeitet
:
RX480
Urgestein
- Mitglied seit
- Feb 13, 2020
- Beiträge
- 1.883
- Bewertungspunkte
- 873
- Punkte
- 114
not every modell has curve values@default, ... but some 6900 should (6900ref Bios)
may be, the other modells without values (0 0 0) have there the problem
Anhänge
I'm using minimal too, you can see if RBAR = SAM is enabled in hwinfo64's main window in video adapters info tree.
Are you sure its enabled in driver as well, not on the device (hardware) level only? I recall it can be enabled in the BIOS, but full adrenaline GUI might show it's not active yet (greyed-out checkbox).
But thanks anyway.
Veii
Veteran
- Mitglied seit
- Okt 24, 2021
- Beiträge
- 184
- Bewertungspunkte
- 212
- Punkte
- 43
#2442
You can not use something that isn't filled. For whatever X reason
The functionality exist and the SKU supports it, tested and confirmed.
Some cards come just empty up to age, go figure. Newest worked on is 6750XT & likely best balanced
StaticVoltageOffset are PLL curves (offsets) additives, ontop of XYZ selected DFLL curve model.
Zero PLL offset, is equally fine to filled in PLL offset
They come post curve and Pre-FIT throttle
They run on PieceWise model, but are just a granular curve shifting droop or an additive, up to users usage
Zuletzt bearbeitet
:
jep, I'm sure gpu-z is showing this, too. the driver interface can only switch this on and off and is not offering additional functionality. meaning when it's switched on in uefi the interface just shows this. iirc a reboot is needed for changing it in driver, therefore it's switching it in uefi, and not in os. it's only meaning your pc is now capable of writing to the full amount of your gpu-memory and not only in 256mb portions.
gpu-z and hwinfo64 shows "disabled" when SAM = r(esizeable)bar is disabled in either uefi or adrenalin interface.
gpu-z is showing this, too. the driver interface can only switch this on and off and is not offering additional functionality. meaning when it's switched on in uefi the interface just shows this. iirc a reboot is needed for changing it in driver, therefore it's switching it in uefi, and not in os. it's only meaning your pc is now capable of writing to the full amount of your gpu-memory and not only in 256mb portions.gpu-z and hwinfo64 shows "disabled" when SAM = r(esizeable)bar is disabled in either uefi or adrenalin interface.
I fully realise this is intended to work with some SKUs (probably 6x50, navi23), and not with the other (particulary those I mentioned), hence the reason why those fields are empty, so I personally still have little comprehension of how useful it is.
Just wanna warn people with a 6800/6900 skus, if you just set Dfll mode to 3 and reboot (not restart64.exe), getting your display back could be troublesome, if your monitor lacks DP or HDMI interface.
Zuletzt bearbeitet
:
Veii
Veteran
- Mitglied seit
- Okt 24, 2021
- Beiträge
- 184
- Bewertungspunkte
- 212
- Punkte
- 43
Yes, 6800XT and 6900XTX are strange and left behind
6900XTXH & 6950 work (Navi 21)
6700, 6750 work (Navi 22)
6600 works (Navi 23)
Its more an AMD oopsy
~ all SKUs support it but it just misses sometimes
Some cards did not get newer vbios either, it takes time
Ones that dont work on stock , just can load an MPT from their same CU SKU and everything should be fine
Or check for bios updates.
EDIT: Or just fill it themselves sooner or later, hopefully sooner
We are in a state of powerplay custom modification and overclocking
A state where you can have multiple device entries & have to mess with registry.
One can expect (hopefully sooner when it's clearly visible) why it would or wouldn't work
A safety cap and auto toggle, would be more delay for a feature we need.
AMD is to blame, and less MPT Team
MorePowerTool is already an advanced book and not for people who don't know how to reinstall windows or boot to savemode to recover their card
Sure a warning wouldn't be bad, but the question resolves itself once all DFLL visibilities came.
The reason this toggle exists, is because the PW-Linear settings made no sense to exist "exactly because" cards were defaulting to DFFL #03 instead of 01
My time of silence expired and goal remains to get some progress done finally, after over a year or so.
I understand researcher's moodset that he was pissed about no progress on Navi since such a long time, and i can share the similar goal.
Soo i pushed hellm to logically add this DFLL selector (after September started), soo people can use PW-Linear in the meantime, till i can talk more about it
There is not one option i would take back, but rather even request for more work on @hellm and remain team
Example:
FlopsPerByte sync values and remain PSM values to be editable.
We'll see what we really need. But removing stuff is the least we should focus on.
We "could" also request a cap on BTC to 100mVMorePowerTool is already an advanced book and not for people who don't know how to reinstall windows or boot to savemode
Sure a warning wouldn't be bad, but the question resolves itself once all DFLL visibilities came.
But where do we go with that. Only more and more delays
Especially since options & functionality are open for change.
If it was for me, i would delay it further till the start of October. When all research between SKUs is done and ground set. To have my piece of credit
But its troublesome with a bigger group forcing everyone to buy Hydra + no support for Intel
And as it remains my part of work to decide what we do with curve and how far we push (no silentness anymore)
// i decided for public approach = let us use this, everyone of us and have a Programm that boots up fast , next to Hydra. Let's finally make some progress
Alternative option would be just to ask for a different UPP on windows , personal tool with Curve GUI approach
// which btw curve-helper dev is quiet, so i'll take that as an "i have no interest to continue this project"
If Hellm wouldn't be fit continuing his work.
I decided for MPT. So please let us keep what we have and expand further
Problem here and there, of course sometimes you'll hardlock drivers and have other side effects
it's fine
Zuletzt bearbeitet
: