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

cra3khead schrieb:
Should I apply extra thermal putty to other areas of the backplate? I'm just being cautious since this is my first time repasting a GPU, and I want to make sure I understand everything beforehand to avoid breaking anything.
yes, only the red areas seems enough

the backplate cools vram + vrm
that´s enough
with a 120er fan at the backplate this should work

delta hotspot depends the most@watt
above 95°C is not really usefull, because throtteling kicks slowly in
-->24/7 lower the watt till <95°C
 
Screenshot 2025-05-24 135311.png

Any idea how I could hit 24k? I'm still waiting for the PTM7950 to arrive, but I don't think it would make that huge of a difference for points; since the max hotspot during a test is 100 degrees.
 
  • Gefällt mir
Reaktionen: Moritz Velten
as I said sometimes, your failure
last weeks the 3dMark was in SALE, .... now to late, = bad decision, don´t bought the opportunity

only with demo=off the temps are good enough
for Runs with fclk 1950...1975...2000@1000...1025...1031...1037mV
and some ds=off in feature control

Vram FT2 with a little bit more Volt?
(my example in TSE see attachment)

What is your minClock, 2500MHz in MCT?
try ULV 1050mV TDV 1056mV together with mooore UVen
MPT= 1112mV, may be with MCT = 1012-->1000...987...981...978mV
 

Anhänge

  • TSE@993ft2.jpg
    TSE@993ft2.jpg
    238,7 KB · Aufrufe: 81
Zuletzt bearbeitet:
  • Gefällt mir
Reaktionen: ApolloX30
cra3khead schrieb:
Any idea how I could hit 24k?
Where do you live - can you put the PC outside for lower air temperatures? If the cooling limits you here, there is not much to gain with all this putty stuff.
 
He lives in Texas.

1748125832127.png


if a dumb Benchmark is sooo important, then only Liquidmetall works around 330W tgp = 390...400W tbp
because above 90°C Hotspot the throttel starts slowly

the 6900 TrioX hasn`t a big Cooler
only 3x92mm fans ?, .... older Torq 4
(This are not the good One´s like SuprimeX)
 
Zuletzt bearbeitet:
may be 6mV more SOC helps
but 2108FT2 is also a good try because less failure correction@high temps
the difference should be null

for example 24/7 I don´t use 2068 but 2066MHz Vram
gives me the same as 2068, ... no negativ effect

I like it to have some headroom.
 

Anhänge

  • zz_TSneu@151W.jpg
    zz_TSneu@151W.jpg
    222,3 KB · Aufrufe: 83
Zuletzt bearbeitet:
  • Gefällt mir
Reaktionen: cra3khead
Yes, I know, you are a one of the cool waterboy´s.
may be there is a first slight throttle in your range
but
the bigger much worse impact@air starts higher

I like your Balkon Temps!
much lower then in Texas
 

Anhänge

  • Balkon3°C.JPG
    Balkon3°C.JPG
    14,3 KB · Aufrufe: 69
Zuletzt bearbeitet:
  • Gefällt mir
Reaktionen: ApolloX30
Is there any anyway to make the effective clockspeed closer to 2660MHz without having it set higher in MCT? My card won't exceed that speed while undervolting.
 
look not only to gfx clock but gfx clock (effective)
ds GFX = off

other values are the tolerances
try 6mV less
Unbenannt.JPG

but
may be UVen is not so easy with less Tolerance

UVen rules over Clock, if you want a higher TS score
the avg. Clock depends more@UVen then maxClock
 
5MHz less maxClock in MCT
and play a little bit with the minClock ... 2500 +/- 5 MHz

Games are very different to TS
you run TS@default but Games are tweakable in Settings +also Tesselation lower per Adrenalin
therefore it would be possible to run a higher MPT Setting in Games with the same Watt
 
Zuletzt bearbeitet:
GerryB schrieb:
probably a other way is also possible, ... set a little change with 25MHz in LD2+3+4 ... 2500-->2525MHz
6900trio.JPG

I used this approach instead of only lowering the tolerance from 48 to 42, and it works much better. This way, I see an increase in my TimeSpy score rather than a drop, and my peak power draw during games is slightly lower.

Can the curve be optimized further for a max clock of 2660MHz in MCT or should I just keep it as it is?
 
try and error
I would think 15MHz more = 6.25mV more works only in the best case, may be
at some point more additional mV are requiered

2660 maxClock (MCT) should be eff. 2600MHz
try LD4 = 2.600 with 0.131250 .... 0.134375 ... 0.137500

btw.
if stable, then as extra test, the other upper value only with 10MHz more with 6.25mV more
LD5=3.100MHz -->2.800 MHz with 0.256250 ... 0.262500
works only up to 2793MHz (MCT)
 
Zuletzt bearbeitet:
  • Gefällt mir
Reaktionen: cra3khead
OK, seems to be stable.
try some MHz more
LD4 = 2.615
LD5 = 2.815

btw.
if stable, then as extra test, LD5 = 2.850 MHz
(don´t know if LD5 has a impact, if the real eff. clock < LD4)
 
Zuletzt bearbeitet:
cra3khead schrieb:
I tried LD4 = 2.625, but it crashed.
OK, stay with LD4 2.620-->2.615
GerryB schrieb:
if stable, then as extra test, LD5 = 2.850 MHz
(don´t know if LD5 has a impact, if the real eff. clock < LD4)
now only more LD5
2.850--> 2.900 ... 2.950 ... 3.000 ... 3.050 .... 3.065 ... 3.075 ... 3.080
may be there is some headroom more in LD5
 

Anhänge

  • 6900_LD.JPG
    6900_LD.JPG
    25,1 KB · Aufrufe: 77
Zuletzt bearbeitet:
Zurück
Oben