Sammelthread Big Navi RX 6700XT/6800(XT)/6900XT Overclocking/Undervolting

[cra3khead]

but you ask also, what is with higher eff.MHz.(x)

Like, are those the absolute best settings for my card with the LD4 optimization, or should I push up the voltage and fmax slightly more?

 

[GerryB]

try only in small steps to go higher, ... 10MHz more with only 6mV more would be nice
may be, at this point also more OFFSET works (MPT to MCT -Volt)
and with powerlimit 98% in MCT, the Temps should be right
and
may be minClock=1473 is little bit more stable then default 500

I dont know how Chill interpolated the Volt in Games with Fps-Limit.

btw.
may be Offset MPT-MCT Volt is in 3dMark more stable then in real heavy Games
if not stable try gaming with MCT 3...6mV more

 

[=dantE=]

Even without pointing out that @GerryB completely is nuts when it comes to optimizing GPU settings on AMD cards, but it's really awesome what he's pushing out in terms of support.
I was until now quiet ... learned a lot from his posts ... i much apprecchiate the support you provide us (nearly) every day.

Thanks Man ...

 

[cra3khead]

@GerryB Why such a big jump in voltage for 10MHz?

Edit:

It crashed while switching to GT2

30 points more; now it's the same as my 1050mv score.

 

[GerryB]

@GerryB Why such a big jump in voltage for 10MHz?

probably works 1/2 6.25mV step

It crashed while switching to GT2

try less Offset MPT to MCT

give a little bit more MCT-Volt, 966 or 967 could be stable
if not
then 2583-->2578MHz

Ergänzung (6. Juni 2024)

Even without pointing out that @GerryB completely is nuts when it comes to optimizing GPU

this 99,9% optimization@cra3khead is very special and works probably only with his good chip!
LD4 2500-->2525 with the same volt is very good
also
LD3 0,089000-->0,087500 means less volt at the same clock (1/4 6.25mV step)
I dont know how exactly the VRM´s can work, but i wonder why AMD use such a 0,089 value.


btw.
I have fun testing 1 little step per day during my morning coffee, and so I can tell cra3khead, what would I do
at this point of setting
but
my 6800nonXT is a little bit different and so not every idea must work on the 6900

 

[cra3khead]

Anhang anzeigen 1491290
probably works 1/2 6.25mV step

LD3 at .0875 isn't stable It's on the edge though; I play games fine but then get a soft crash after an hour; I'm going back to the previous settings with 1056 at 2573 with prior optimized curve settings; maybe my card is tapped out unfortunately.

I have fun testing 1 little step per day during my morning coffee, and so I can tell cra3khead, what would I do
at this point of setting

I do the same

 

[GerryB]

I do the same

Ok,
with your older stable setting try
a little bit less Aging Value

 

[cra3khead]

with your older stable setting try
a little bit less Aging Value

Anhang anzeigen 1491293

No good. Would LD3 at .08825 work, or would it be pointless because it's 1/8th step? Or how about 2315MHz at .0890?

 

[GerryB]

If the run crashs in GT1 or GT2 at the satellite, then try 5MHz less maxGFX-Clock.

I would stay with LD3 at 2300 with the default.
may be 2315...2325@0,0937500, but no big impact to expect, same Score?
or 2310 with 0,0906250

 

[cra3khead]

LD3 at .0875 isn't stable It's on the edge though; I play games fine but then get a soft crash after an hour

Weird when I went back to LD3 at default .0890 in the same game, it crashed within 10 minutes, @GerryB How's it possible that the undervolt is more stable?

 

[GerryB]

may be, undervolt is a little bit less eff.clock in the same game scene

If you are not sure, whats the reason, try MCT with 1mV ... 2mV more.
In GAMES its better to use a little bit more MCT-Volt then in 3dMark.
(more clock changes dip+spike)

 

[cra3khead]

may be, undervolt is a little bit less eff.clock in the same game scene

I think the optimized curve is only stable in certain games; I went back to my standard 1050mv profile with no issues. I guess it was just running well previously for MW3.

 

[GerryB]

your 1050mV-setting is already very nice with that high max clock per volt
(with MCT powerlimit 98% = 275W there should be in MW3 nearly the same fps)

 

[cra3khead]

your 1050mV-setting is already very nice with that high max clock per volt

You think that since 2525 LD4 kind of works for 2573/1056, should I just set LD4 to 2515 on the 1050mv profile but not touch any of the other settings? Would that allow for the extra boost I got with the curve optimization profile, but then also be more stable?

 

[GerryB]

should I just set LD4 to 2515 on the 1050mv profile but not touch any of the other settings?

may be also 2525MHz works, because the real eff.MHz+Volt are lower
and
as second step LD3= 0,090625 Volt (if 0,087500 Volt is not stable enough)
and
as third step LD3 with 10MHz more
(in this kind a crazy Test with my 6800 see attachment, in PR@RT goes more Clock@lessVolt then in Games@Raster,
in RasterGames 15MHz less with more Adrenalin-Volt)

I think, in some games the change in tolerance 48-->43 was to much with your setting/chip.
for safety back to default 48

 

[cra3khead]

as second step LD3= 0,0878125 Volt (if 0,087500 Volt is not stable enough)

There's not enough room for the last digit of 5 to fit, though.

I think, in some games the change in tolerance 48-->43 was to much with your setting/chip.
for safety back to default 48

If I change it back to 48, do I have to change the vmin values back to the default of 825 and 775 for ulv min, or are they unrelated?

With the new profile, should I keep the LD2 default? I don't think it's worth changing since my card barely goes to that low of a clock in any game I play.

Weird

57 more points over the standard 1050mv profile. I think the LD4 just causes instability since it doesn't hit over 2.5Ghz anyway. My highest score at default tdp (281 Watts) though !

 

[GerryB]

There's not enough room for the last digit of 5 to fit, though.

= no problem, the tool+adrenalin rounds intern up

If I change it back to 48, do I have to change the vmin values back to the default of 825 and 775 for ulv min, or are they unrelated?

no, the idea with with minGFX 818 was independent (x)
only the requiered MCT-volt could a little bit affected

Anhang anzeigen 1491752

57 more points over the standard 1050mv profile. I think the LD4 just causes instability since it doesn't hit over 2.5Ghz anyway. My highest score at default tdp (281 Watts) though !

nice try with LD2, ... should this be more stable with the LD3-UV?
If this works in games together with minGFX 825mV, then stay so.
(x) may be minGFX 818mV is too less with MPT1050+MCT950, needs MCT 951... 954 ... 956mV

 

[cra3khead]

nice try with LD2, ... should this be more stable with the LD3-UV?
If this works in games together with minGFX 825mV, then stay so.

I wish I could push Fmax up more though; I tried the 2535/0.1062500 option you suggested a while back, but that didn't work. Maybe I'll just run 1050mv in the summer, and in the winter hopefully it should be cold enough to run 1125mv at 2.6Ghz without as much of a heat issue.

 

[GerryB]

with the 1056mV and 1062mV-Setting you can try:

If LD3 0,087500mV is not enough in Games, then 0,090625mV, ... should be the same as default 0,089000
or 1/4 x 6.25mV more (I dont really believe, that 1/4 steps work, but 87,5mV+6,25/2 looks good)

DcBtcGb 6mV-->12mV is a little bit slower, but safer

 

[cra3khead]

Anhang anzeigen 1491752

This profile is slightly more stable than the previous 1056mv with the curve optimization, but it still crashes; this time in, like, 2-3 hours of gameplay rather than 1 hour. Don't know what could be causing that.

DcBtcGb 6mV-->12mV is a little bit slower, but safer

Could this help stabilize my 1050mv profile?

Probably not