6800/6900 Overclock Results

Did a run this AM. From my Wattman settings above, I set the fan to max out at 75% instead of 100%, and lowered min clock from 2575 to 2500. Graphics score increased 150+ points.

MPT still has no effect on my card. It really would be nice if they would release RBE with support for Big Navi.
 

Attachments

  • 20200.JPG
    20200.JPG
    291.2 KB · Views: 0
In testing with TD2 this morning I've noticed that the voltage still ends up at or near 1150mv while in-game playing a control point. So despite being set for 1080mv in the driver it doesn't really maintain voltage that low in practice while under load. This also explains why junction temps seem sort of high to me at 92-94c even when the fans are spun up 75%. It's using power (320-330w), hence producing heat.

I'd like to lower the temps a bit and pull back so I can cut the fans down a notch if possible. Doing so would probably push me under 20k timespy though.
You should think of the voltage you set as a percent offset rather than absolute. Essentially the card would have gone to 1150 at a lower frequency without your adjustment. It’s also worth noting that most games (outside of unreal engine) won’t load up your card nearly as hard as timespy. For example, settings that hit consistent 360w in Timespy for me average about 300-310w in Cyberpunk. See how the temps look for you in actual games.
No but I will try it. AB is much easier to use than Wattman. Wattman is dumb.
I really don’t see the advantage of afterburner over wattman. The vertical ellipses in the top right above where you select “manual” will let you save your profile and reload easily after a crash.
Af
 
Did a run this AM. From my Wattman settings above, I set the fan to max out at 75% instead of 100%, and lowered min clock from 2575 to 2500. Graphics score increased 150+ points.

MPT still has no effect on my card. It really would be nice if they would release RBE with support for Big Navi.
Thats a great score!
I havent tried MPT yet, but I am using reference design so im kinda 50/50 on it till I can get a waterblock.
 
I've got a 6800 with a 6900XT cooler on it that I run every day at 975mv with the wattman GPU slider at 2275MHz and stock VRAM clocks.

I can bump the core to the 1025mv max it allows in wattman and run the GPU slider at 2600MHz with VRAM at 2100 with no issues. I'll post some numbers up later.
 
I'm really considering giving this a go on my 6900XT. My only hesitation is that the last time I did something similar with a RX5700 it ended up dying on me very quickly. I had it flashed to a 5700XT and then did the power play mod to increase it's TDP and the card lasted maybe a week before giving up the ghost.
 
I wouldn’t recommend MPT without liquid cooling, these get hot enough on air as it is.
 
I wouldn’t recommend MPT without liquid cooling, these get hot enough on air as it is.

currently on water I hit around 50c core 70c junction with 2550-2650 with +15% and 1130mv. Think I have anything to gain via MPT?
 
I wouldn’t recommend MPT without liquid cooling, these get hot enough on air as it is.
Yeah that was my assessment as well. Currently the card (6900XT Nitro+) is on it's stock cooler and I'm in no hurry to kill a $1200+ nearly impossible to find video card at the moment. I originally wanted a card with a reference design just to make the possibility of watercooling it easier but I was not able to secure a reference board unfortunately.
 
Yeah that was my assessment as well. Currently the card (6900XT Nitro+) is on it's stock cooler and I'm in no hurry to kill a $1200+ nearly impossible to find video card at the moment. I originally wanted a card with a reference design just to make the possibility of watercooling it easier but I was not able to secure a reference board unfortunately.
If only I had known. I have a sapphire reference 6900xt and just bought a nitro+ version. Maybe could have helped each other out. Likely going to just sell the reference card now.
 
For the 6800xt and 6900xt, just going with the auto undervolt is a small performance increase that's a bigger increase in performance than Rage mode and you get a bit lower thermals; this is a great solution if you just want to do a one click performance increase that's worth a few percent.

I haven't really found the increases in performance worth the increase in power consumption and thermals honestly, though that doesn't really matter much if you went to water with the intent of overclocking.
 
For the 6800xt and 6900xt, just going with the auto undervolt is a small performance increase that's a bigger increase in performance than Rage mode and you get a bit lower thermals; this is a great solution if you just want to do a one click performance increase that's worth a few percent.

I haven't really found the increases in performance worth the increase in power consumption and thermals honestly, though that doesn't really matter much if you went to water with the intent of overclocking.
It was meaningful for me in Cyberpunk at 4K high, with my OC I never drop below 50FPS which keeps me up in the safe zone of my freesync range. I concur more generally though.
If only I had known. I have a sapphire reference 6900xt and just bought a nitro+ version. Maybe could have helped each other out. Likely going to just sell the reference card now.
You might find a buyer in this thread. :)
currently on water I hit around 50c core 70c junction with 2550-2650 with +15% and 1130mv. Think I have anything to gain via MPT?
Yeah, I gained about 150mhz peak speed with MPT (went from setting at 2650 to 2800). My core temps went from 40 to 45C with the extra power, you're starting at a higher temp so maybe a little less headroom. Now, whether you consider a 20% increase in power draw for a 5% higher timespy score worth it probably depends on the user. Nvidia certainly seems to be finding plenty who buy that logic. ;)
 
It’s also worth noting that most games (outside of unreal engine) won’t load up your card nearly as hard as timespy.
From what I've seen with TD2, it loads the card or at least the core anyway higher than timespy. In timespy my max junction temp is in the high 80s and with TD2 it hits low 90s.
 
From what I've seen with TD2, it loads the card or at least the core anyway higher than timespy. In timespy my max junction temp is in the high 80s and with TD2 it hits low 90s.
Interesting, I've not played anything with that engine before.

For games that render way above my refresh rate, I just pull the power slider back in wattman.
 
Played around with it a little bit tonight and from what I can tell there is a pretty narrow range where clock settings and voltage are stable. If I push voltage down from where it's at now (1080) it becomes unstable. I can reduce clock speeds but if I lower voltage along with clocks I still run into issues. If I increase voltage and try for higher clock speeds like 2700/2600, it's unstable too.

It seems like there is this spot where it seems to work okay but if I depart from that, my sample has problems. I'll tinker some more with it over the weekend but I don't think I can get clocks any higher than what I'm at now (2660/2560). I may be tapped out on air in the low 20k range. Oh well I guess. I'm fine with that result.

EDIT: Left the voltage at 1080 where it was and just reduced clocks to 2500/2400. That resulted in a lockup. Kind of counter-intuitive really. You would think that if it is stable at 2660/2560 and 1080mv, then why would it be unstable simply by reducing clocks?
 
Last edited:
Played around with it a little bit tonight and from what I can tell there is a pretty narrow range where clock settings and voltage are stable. If I push voltage down from where it's at now (1080) it becomes unstable. I can reduce clock speeds but if I lower voltage along with clocks I still run into issues. If I increase voltage and try for higher clock speeds like 2700/2600, it's unstable too.

It seems like there is this spot where it seems to work okay but if I depart from that, my sample has problems. I'll tinker some more with it over the weekend but I don't think I can get clocks any higher than what I'm at now (2660/2560). I may be tapped out on air in the low 20k range. Oh well I guess. I'm fine with that result.
That seems to be the sweet spot of both my cards as well. 2550/2650. Both clock almost identical and score about the same, low 20k in Timespy.
 
That seems to be the sweet spot of both my cards as well. 2550/2650. Both clock almost identical and score about the same, low 20k in Timespy.
Same for me before I played with the power limit. Seems like a common balance point at the 300W limit. Still nice that its a solid 350-400mhz above advertised boost clock!
 
I tried lowering the voltage to 1035 and got a CTD in the demo. 1040mV finished the run and got me a whole 3 more points to 20203. I bought a key so I could skip the demo and did a run at 1040mV. For my money I got 62 more points for 20265.

Edit: I linked my Steam account to a 3DMark account I had forgotten about. Brought back some memories.:) Also, did a run of Port Royal.
 

Attachments

  • 20265.JPG
    20265.JPG
    300.1 KB · Views: 0
  • History.JPG
    History.JPG
    171 KB · Views: 0
  • PR10739.JPG
    PR10739.JPG
    270.2 KB · Views: 0
Last edited:
So is the consensus to avoid mpt if on air? I don't want to send the 6900 to an early grave.
 
So is the consensus to avoid mpt if on air? I don't want to send the 6900 to an early grave.
Not sure about the consensus, but for playing around with, I think it is okay if you do not go crazy. It you boost yourself an extra 10-20W maybe not a big deal as long as thermals are in check. But I would not run the higher limits 24/7 on air personally. The marginal performance gain is likely not worth the risk.
 
If using afterburner my timespy score was down about 130 points. It may have something to do with being unable to set minimum clocks. I ended up uninstalling it. Also the fan curve in afterburner keys off the GPU temp and not the junction temp, whereas in wattman the curve points seem to go off junction temp.

I really am not comfortable pushing mine any further on air as it is seeing spikes up to 94c junction temps in game.
 
First post here. So be gentle with this virgin lol. I don't understand everything yet but I try my best,
Was googling where to find the rx6000 enthousiasts and stumbled across this forum. :)

This is my first build too so first time trying to oc a gpu. First I read up all kinds of info on forums, watched a shitload of youtbe vids. And now I'm here to share my first tries.


GPU is XFX Speedster Merc 319 6800XT. No MPT/Air Only.
Getting stable graphics back2back scores in Time Spy between 19700 and 19770 with:

Max/Min 2650/2470
VRAM 2150
Voltage 1030
Fans 75%


Wattman settings for my highscore run which shows some dips but still managed to get a valid result:

Max/Min 2675/2475
VRAM 2150
Voltage 1030
Fans 100%
 

Attachments

  • Time Spy 31-01-2021.PNG
    Time Spy 31-01-2021.PNG
    1,007 KB · Views: 0
  • Time Spy 31-01-2021_Stable.PNG
    Time Spy 31-01-2021_Stable.PNG
    1,002.5 KB · Views: 0
First post here. So be gentle with this virgin lol. I don't understand everything yet but I try my best,
Was googling where to find the rx6000 enthousiasts and stumbled across this forum. :)

This is my first build too so first time trying to oc a gpu. First I read up all kinds of info on forums, watched a shitload of youtbe vids. And now I'm here to share my first tries.


GPU is XFX Speedster Merc 319 6800XT. No MPT/Air Only.
Getting stable graphics back2back scores in Time Spy between 19700 and 19770 with:

Max/Min 2650/2470
VRAM 2150
Voltage 1030
Fans 75%


Wattman settings for my highscore run which shows some dips but still managed to get a valid result:

Max/Min 2675/2475
VRAM 2150
Voltage 1030
Fans 100%
That’s a solid OC. With your voltage that low, you may have room under the power limit (ensure it’s at +15%) to increase voltage and go higher. I’d be curious on your results with these settings:

2650@1050

And
2700@ 1070
 
2650@1050 was around the same results as my stable 2650@1030 clock. But it obviously ran more hot.
2700@1070 was continously crashing.
Got stable at 2660@1035 with graphics scores between 19820 and 19870 but it shaved almost a whopping 1000 points off my cpu score in every run (around 8k instead of 9k)
 
2650@1050 was around the same results as my stable 2650@1030 clock. But it obviously ran more hot.
2700@1070 was continously crashing.
Got stable at 2660@1035 with graphics scores between 19820 and 19870 but it shaved almost a whopping 1000 points off my cpu score in every run (around 8k instead of 9k)
Yeah, seems like you've got a well dialed in result then. If you're not seeing more score with more voltage you're at a good inflection. Really nice chip.

2700 was a bit of a stretch based on how well you were clocking at 1030. I think you really need water temps to get to 2.7 or higher on these.

The lower CPU score is interesting, indicates you had something borderline unstable.

Only two other things come to mind to check:
  1. Look in GPU-Z to see whether you're running at 8x or 16x on the pci-e link. I had to change the card configuration around in my X470 (and leave the second x16 slot empty) to avoid the board bifurcating it down to an 8x link. That was worth 200-300 points in timespy and might get you into 20k land.
  2. Ensure your memory is on "fast timings" and then drop the setting to 2130 vs your 2150. Both my 6800XT and 6900XT scored higher with the slightly slower ram, indicating they were doing error correction at the higher clock.
 
Yeah, seems like you've got a well dialed in result then. If you're not seeing more score with more voltage you're at a good inflection. Really nice chip.

2700 was a bit of a stretch based on how well you were clocking at 1030. I think you really need water temps to get to 2.7 or higher on these.

The lower CPU score is interesting, indicates you had something borderline unstable.

Only two other things come to mind to check:
  1. Look in GPU-Z to see whether you're running at 8x or 16x on the pci-e link. I had to change the card configuration around in my X470 (and leave the second x16 slot empty) to avoid the board bifurcating it down to an 8x link. That was worth 200-300 points in timespy and might get you into 20k land.
  2. Ensure your memory is on "fast timings" and then drop the setting to 2130 vs your 2150. Both my 6800XT and 6900XT scored higher with the slightly slower ram, indicating they were doing error correction at the higher clock.
Got stable clocks in from voltage range 1030 to 1045 with averages of 19700-19800 timespy graphics score while maintaining stable cpu scores now. pcie link is x16. mem timing was on fast timing at all times and the 2150 to 2130 vram made things maybe a little more stable at the higher clocks but no performance jump in scores here.
Really appreciate the feedback and your explanation. And yes, someday have to waterboard and fully drown this thing indeed to crank out the extra juice.
 
Got stable clocks in from voltage range 1030 to 1045 with averages of 19700-19800 timespy graphics score while maintaining stable cpu scores now. pcie link is x16. mem timing was on fast timing at all times and the 2150 to 2130 vram made things maybe a little more stable at the higher clocks but no performance jump in scores here.
Really appreciate the feedback and your explanation. And yes, someday have to waterboard and fully drown this thing indeed to crank out the extra juice.
your card has one of the top air coolers out of the bunch. best card xfx has ever made to date
 
Thought I'd test AMD's claims of 2X performance/watt. With the power target of 150 watts here. These settings run CP2077 on all medium settings at an average of 140 watts well above my monitors 60Hz refresh.
 

Attachments

  • 150watt setting.png
    150watt setting.png
    598 KB · Views: 0
  • 150watts or less.png
    150watts or less.png
    664.3 KB · Views: 0
Well, setting a minimum GPU clock of 2400Hz instead of the default picked me up another 200 or so points in Timespy. Can't go much below 1100mV on air. Around 1084mV gets artifacting in some games so I've raised it up a bit. I'll have to try VRAM tuning and fan profiles next I guess, although I don't think I'll really go away from the default fan settings except for epeen points because it's so quiet right now.

Still wish my 5800x scored higher on the CPU part of the benchmark after seeing people with 10900k's getting double my score.

3dmark results.png
 
With pbo enabled and ryzen high or ultimate performance power plan my brother gets average 11500 on cpu score with 5800x and a 3080 suprim x.
With my 5600x and 6800xt I get around 8800 average cpu score (pbo enabled and high performance power plan)
Both builds use arctic liquid freezer II 240 rev 3 offset mount.
 
Well, setting a minimum GPU clock of 2400Hz instead of the default picked me up another 200 or so points in Timespy. Can't go much below 1100mV on air. Around 1084mV gets artifacting in some games so I've raised it up a bit. I'll have to try VRAM tuning and fan profiles next I guess, although I don't think I'll really go away from the default fan settings except for epeen points because it's so quiet right now.

Still wish my 5800x scored higher on the CPU part of the benchmark after seeing people with 10900k's getting double my score.

View attachment 325165
Have you tried turning SMT off? I know the high core count cpus get a 1k bump to CPU score with it off. Not sure if 8 cores will see an improvement but worth a try of you haven't done this yet. I would like to know if you see an improvement.

My 3600 at 4.4ghz allcore OC with smt on gets about 7.5k and with it off it drops to 5.5k. So it does not help with 6 core CPUs.
 
With pbo enabled and ryzen high or ultimate performance power plan my brother gets average 11500 on cpu score with 5800x and a 3080 suprim x.
With my 5600x and 6800xt I get around 8800 average cpu score (pbo enabled and high performance power plan)
Both builds use arctic liquid freezer II 240 rev 3 offset mount.
Ryzen power plan was removed for the 5000 series in the chipset drivers. Windows is able to handle the power plan for the 5000 series as intended. Quoted from AMD support forum

"Ryzen 5000 series CPUs do not require a special Ryzen Power Profile, so it is not installed with Chipset driver package. Please use the Windows Balanced Power Profile for these procesors. Previous generation Ryzen processors will continue to use the Ryzen Balanced Power Profile for optimal performance and this is included in the Chipset driver package.


Starting with AMD Chipset Driver 02.10.13.408, the processor power can be adjusted via Performance and Energy slider on systems running Windows 10 2004 and later."
 
Well, setting a minimum GPU clock of 2400Hz instead of the default picked me up another 200 or so points in Timespy. Can't go much below 1100mV on air. Around 1084mV gets artifacting in some games so I've raised it up a bit. I'll have to try VRAM tuning and fan profiles next I guess, although I don't think I'll really go away from the default fan settings except for epeen points because it's so quiet right now.

Still wish my 5800x scored higher on the CPU part of the benchmark after seeing people with 10900k's getting double my score.

View attachment 325165

That is pretty strange, are you running the latest chipset driver with the Windows power plan set to balanced? My 5800X on a Noctua NH-U12A with all default BIOS settings on a B550 Taichi scores ~12000 in CPU.


1612277986458.png





Bumping up the the GPU clock slider to 2600MHz and VRAM slider to 2150MHz

1612279814247.png
 
Last edited:
Didn’t know this will check again.
I’m a total noob after all :D
Thanks for the reply guys!
 
I’m entirely new to the AMD over clocking with their software, so forgive me, but could I have been so unlucky with the Red Devil?

Tried to lower voltage based on things I’ve been reading along with raising the power limit +15, not even touching shader/mem clocks. Started conservatively, eventually dropping to 1070mV. However, it doesn’t matter because once I load up my Control save and assets load in game, it crashes with device removed errors at anything but stock it seems. Also, is it really only auto-clock, auto-mem, or auto-voltage? Or do they change the other parameters as well?

I’m using Control just because I made a choice of a single game at the time with my time constraints (I had a shortcut on my desktop >.>). I do not use benchmark tools and will not. I’m gonna try to keep testing more games this evening if I don’t pass out after the kids go down for bed.

For now I’m at auto over clock (I think it was 2.5 GHz something close) and just enjoying the card and boost over the 2080 Ti. And yes, OC bios is selected.
 
Have you tried turning SMT off? I know the high core count cpus get a 1k bump to CPU score with it off. Not sure if 8 cores will see an improvement but worth a try of you haven't done this yet. I would like to know if you see an improvement.

My 3600 at 4.4ghz allcore OC with smt on gets about 7.5k and with it off it drops to 5.5k. So it does not help with 6 core CPUs.
I haven't tried, I read some reviews on it and it seemed like turning it off was a wash, helping in some things and hurting in others so I didn't bother. My temps are relatively high though so I've been meaning to repaste the CPU and see if that helps.

I checked the power plan and I had it on high performance still, but now there's a Ryzen High Performance mode listed so I'll switch to that and retry.
 
Back
Top