6800/6900 Overclock Results

Do you have your memory at the max section of the slider?
I have it set at 2144 MHz. Anything above this isn't stable. I can't enable 'fast timing' (hard crash) so I don't know how much it affects the score.
 
I upped power just a bit on my 6800XT to get the 22k score. 298 watts @ max draw. It's an AMD reference black edition card, and while the fans are....not quiet....at those settings the temps are within reason. I don't keep my card pegged at that extreme OC, but I'll game all day with a pretty high OC on the card with zero issues.
I'm at 365 w using MPT but can't go higher as I'm hitting 100* junction temps (on air). Spent plenty of time adjusting GPU and case fans and can't seem to improve temps.
 
Last edited:
I have it set at 2144 MHz. Anything above this isn't stable. I can't enable 'fast timing' (hard crash) so I don't know how much it affects the score.

While I can set mine to 2150, it doesn't help. I get best results at or slightly below 2130.
 
hose of you hitting 21-23,XXX TS scores, are y'all running XTXH-spec cards or cranking up the power limit to 400ish watts (so presumably under water)? Or have driver updates helped since early in the year when we were all getting 20,XXX scores?

My 20.3k score from a few pages back seemed about mid pack at the time (Feb or so).

My 23,411 is with power limit at 430w and under water on an XTXH card.

I think it is very possible to get high scores with lower power and non XTXH chips if you are lucky and get a chip that doesn't mind being seriously under-volted, but that was not mine. It became unstable as soon as I tried to lower voltage below like 1150mv
 
I'm at 365 w using MPT but can't go higher as I'm hitting 100* junction temps (on air). Spent plenty of time adjusting GPU and case fans and can't seem to improve temps.

I have a huge undervolt, I actually didn't save my exact settings from my best run, but it was @ 1050mV or so. Playing with voltage is the where most dividends came from. My card can hit 2.8ghz but will thermal throtlle, but it'll run 2.5-2.6ghz no problem on a full benchmark without thermal throttle and that's where the big score numbers come in.

Edit: Ran a couple quick benches to show settings/temps, don't really have time to push it today but a good example. Note: Purchased a 12700K and moved to Win 11 since the previous benches so that's altering results some I'm sure with a new motherboard/OS.

1640381309604.png

1640381264383.png
 
Last edited:
I have a huge undervolt, I actually didn't save my exact settings from my best run, but it was @ 1050mV or so. Playing with voltage is the where most dividends came from. My card can hit 2.8ghz but will thermal throtlle, but it'll run 2.5-2.6ghz no problem on a full benchmark without thermal throttle and that's where the big score numbers come in.

Edit: Ran a couple quick benches to show settings/temps, don't really have time to push it today but a good example. Note: Purchased a 12700K and moved to Win 11 since the previous benches so that's altering results some I'm sure with a new motherboard/OS.

View attachment 425096
View attachment 425095

My scores went down with 11.
 
My 23,411 is with power limit at 430w and under water on an XTXH card.

I think it is very possible to get high scores with lower power and non XTXH chips if you are lucky and get a chip that doesn't mind being seriously under-volted, but that was not mine. It became unstable as soon as I tried to lower voltage below like 1150mv
So, I updated the drivers, reconfigured the case fans, then set the mem clock to 2,130 as thecold suggested. With the core clock at 2,600 (50ish MHz less than the runs in Feb) I ended up at 21.9k. Junction temps in the mid 80's, or about 5* cooler than before the fan reconfig.

Noticed that the clock wasn't boosting past about 2,400 so I set the max power to 345w (same core clock and mem clock settings) with MPT and reran, getting 22.2k this time around. Average clock was right at 2,500 with the increased power. Junction temps in the high 80s, just touching 90*. About 8-9* cooler than prior.

I can see getting your score under water. I think this thing is about tapped out on air. Interesting how my score improved nearly 2,000 points since Feb based only on (1) updated drivers, (2) clocks slightly reduced, (3) enabling 'fast timing' on memory, and (4) lower junction temps. I tested the effects of enabling and disabling the 'fast timing' and enabling it adds about 150 pts to the score.
 

Attachments

  • Best_score_current.png
    Best_score_current.png
    739.7 KB · Views: 0
  • Best_score_currentwMPT.png
    Best_score_currentwMPT.png
    742.1 KB · Views: 0
Last edited:
Has anyone tried repasting their reference 6800/6900? I'm thinking about it on my ref 6800 xt black to try and bring the junction temp down for OC runs. Normal gaming junction never gets above 72c which is really good. But trying to get a good Time Spy score is becoming an issue. haha
 
Has anyone tried repasting their reference 6800/6900? I'm thinking about it on my ref 6800 xt black to try and bring the junction temp down for OC runs. Normal gaming junction never gets above 72c which is really good. But trying to get a good Time Spy score is becoming an issue. haha

I took the reference backplate off, but from what I've seen repasting is not only a PITA but not huge improvements. My temps on the 6800XT BE haven't really been an issue, though, with some aggressive fan curves.
 
I took the reference backplate off, but from what I've seen repasting is not only a PITA but not huge improvements. My temps on the 6800XT BE haven't really been an issue, though, with some aggressive fan curves.
You've got a good sample though. My gaming temps are fine. But hot spot in time spy will get to 110c easily, even 100% fan. But It's also pulling 350watts or more. Best score was 20,300. The thermal pad does a pretty good job, and once you break that there is no going back. So you're either stuck with trying to find another pad, or get lucky with paste. I need a second card for science...
 
I think if you do washers to increase mounting pressure and a great paste you MIGHT be able to get 10c, maybe. You're still pushing 350 watts through an OEM heatsink. The heatsink is not hard to remove on this card, give it a try.

On water at 350w my hotspot hits around 70c.
 
You've got a good sample though. My gaming temps are fine. But hot spot in time spy will get to 110c easily, even 100% fan. But It's also pulling 350watts or more. Best score was 20,300. The thermal pad does a pretty good job, and once you break that there is no going back. So you're either stuck with trying to find another pad, or get lucky with paste. I need a second card for science...

No other issues pushing 350 watts through? I've been hesitant to push past 300 watts just due to lack of confidence in the reference power delivery.
 
Make a chart someone is willing to maintain with 6800/6900 3dmark results like the 3090 thread, I'd love to see what people are pulling on higher ghz insanity with the better binned silicon.
 
What should I be aiming for on a Gigabyte waterforce 6900xt power and clocks wise?
 
Hi, I got a reference 6900XT last October after 6 months of trying my luck in the weekly Thursday drop on AMD.com, I almost fell of my chair that I finally was able to get from the cue in the shop to place an order ... had to lay down for 5min to recover from that.

Happy I stumbled on this tread, had fun reading it and thought I'd also give it a try with the pointers read here:
But first try gave me 12411 graphics score ? On second try, luckily, I spotted the fps at a steady 75 ... yeah, ok, I had Radeon Chill still on :D
With that out of the way, here are some scores and findings with different settings (but with power limit always +15% and SAM enabled):
  • 20660 (1175mV, VRAM default)
  • 20780 (1175mV, VRAM fast timing @2100)
  • 20959 (1175mV, VRAM fast timing @2150)
  • 21100 (1160mV, VRAM fast timing @2150)
  • 21171 (1150mV, VRAM fast timing @2150)
  • 21239 (1140mV, VRAM fast timing @2150)
  • 21491 (1140mV, VRAM fast timing @2150, SMT off)
  • 21512 (1140mV, VRAM fast timing @2150, SMT on, RAM OC)
  • 21499 (1140mV, VRAM fast timing @2150, SMT off, RAM OC)
Power draw seemed to be max 293W, and max GPU clock seemed to hoover around 2360-2400, with temps around 70°C.
Tried 1130mV, but that would cause the second graphics test to fail.
SMT off (Ryzen 5900X) the first time gave me a boost off 250 points, but when overclocking RAM, there seemed to be no difference anymore in graphics score when turning on/off SMT ?

Original plan was to put an EK classic WB on it, but didn't get to it yet ... I just really like how the reference card looks :D, and it is also very quiet even under load, it hangs at the wall at arms length in an open air case and it never bothers me (... in comparison, my old Vega64 reference, that HAD to be put under water, because it sounded like a vacuum cleaner).
 

Attachments

  • 21512.png
    21512.png
    614 KB · Views: 0
Any of these AMD Reference 6900XT? Trying to get an idea of what to expect.
 
Last edited:
Mine is an XFX merc 319
 

Attachments

  • Screenshot_2021-10-11 I scored 19 542 in Time Spy.png
    Screenshot_2021-10-11 I scored 19 542 in Time Spy.png
    434.1 KB · Views: 1
Do you have your memory at the max section of the slider?

You don't necessarily want to max out the RAM. For me anything over 2025-2030 actually resulted in a DECREASE in performance/score. 2025-2030 was the sweet spot for me. Go a little beyond it and performance starts to go down, go far beyond it and we have crashes.
 
Tried looking for yours on here and didn't find it.
Just reran it. Last year I got just over 16k on the total score early on and I think around 18-19k on graphics after playing around with overclocks. Put a better cooler on my 5800x and overclocked my GPU to:
2704mhz core
stock memory timing & speed
1134mv core voltage
+15% power limit
Increased fan speed

Just got 16648 total score, 19266 graphics score, and 9406 CPU score.

system specs
AMD 5800X processor
6900XT GPU (SAM enabled)
3600MHz RAM (32GB @ 16-19-19-39)
Resolution: 1440p
 

Attachments

  • timespy.png
    timespy.png
    719 KB · Views: 1
Tried looking for yours on here and didn't find it.

Post 300 is my score for a AMD reference 6900xt. It ran pretty hot, so it's now under water but it did not improve my scores much, but it did drastically drop the temps.
 
2704mhz core
stock memory timing & speed
1134mv core voltage
+15% power limit
Increased fan speed
That score looks on the low side ... what else could be at play here, CPU/RAM, VRAM ?
You got very high clock speed (2500+), but lower graphics score (19266) ?

Gideon's reference 6900XT and mine look to have very similar results:
Both score around 21250 (without OC or SMT changes) and have a GPU clock speed around 2250-2300 when looking at the 3DMark graph...
 
Could it be that the min-max frequency setting has some effect on the score ?
I've seen ppl mentioning using 'min 2500 - max 2600' or similar ... during my tests, I didn't touch those sliders yet, so they were at the default (min 500 - max 2484 ?).

I'm not really sure what they are for, actually. I can imagine by setting a lower max you can limit power draw or lower temps, or gain stability ... unless you can have more power delivered and have good cooling, then it makes sense going beyond :)

I ran 3dMark again a couple of times with similar values to 2500-2600 and got results that were a lot worse by only changing those min-max settings:
  • 21239 (min 500 - max 2496)
  • 18906 (min 2511 - max 2694)
To check to difference, I overlaid the 3DMark graphs of the to runs and the higher score came from 'higher FPS achieved with lower clocks' ?
(grey one is the higher score)
 

Attachments

  • low vs high 02.png
    low vs high 02.png
    25.8 KB · Views: 0
  • low vs high 01.png
    low vs high 01.png
    137 KB · Views: 0
That score looks on the low side ... what else could be at play here, CPU/RAM, VRAM ?
You got very high clock speed (2500+), but lower graphics score (19266) ?

Gideon's reference 6900XT and mine look to have very similar results:
Both score around 21250 (without OC or SMT changes) and have a GPU clock speed around 2250-2300 when looking at the 3DMark graph...
Yeah, not sure. Always scored lower even when i spent more time tweaking it. Tried raising the clock speed more but it's crashing. I didn't think to change the min value this time, it's still set at the default 500mhz but I think I remember increasing it before.
 
Yeah, not sure. Always scored lower even when i spent more time tweaking it. Tried raising the clock speed more but it's crashing. I didn't think to change the min value this time, it's still set at the default 500mhz but I think I remember increasing it before.
Interesting ... I played around a little bit more and in my case, the max slider doesn't seem to have much effect, but as soon as I put the min slider around 2500 (so at least 100 higher than my 6900 normally does), then the clock speed effectively raises, but at the same time, the FPS is lower and score drops by 2000-3000 points (lol, just got 17821 :D by setting 'min 2530 - max 2705)
 
Interesting ... I played around a little bit more and in my case, the max slider doesn't seem to have much effect, but as soon as I put the min slider around 2500 (so at least 100 higher than my 6900 normally does), then the clock speed effectively raises, but at the same time, the FPS is lower and score drops by 2000-3000 points (lol, just got 17821 :D by setting 'min 2530 - max 2705)


Ok, just retried. I seem to remember someone saying setting the min core clock higher but not too high works. I set the min core clock to 1100mhz and kept everything else basically the same and got 21023 on graphics and 17,592 total.
 
Ok, just retried. I seem to remember someone saying setting the min core clock higher but not too high works. I set the min core clock to 1100mhz and kept everything else basically the same and got 21023 on graphics and 17,592 total.
Ah, that's a good improvement ! And thanks for confirming.

So while in 3DMark, it has a negative effect, I've played around a little more with this min-max setting with 2 in-game benchmarks to see what the effect would be there ... they might not be representative of real performance, but easy to use: Far Cry 6 and Deux Ex: Mankind Divided.
So, also here, the GPU clocks speed is a lot higher with high min-max (according to GPU-Z), but there was no real performance difference.

Here is an example of Far Cry 6 at 'high' preset:
The clock speed graph is very different, with high min-max at a steady 2608, and the +/- default setting at much lower and variable clock speed ...
All other graphs look very similar, and the performance was the same: +/- 83 average fps, +/- 65 low fps.

Is there be a situation/benchmark where this higher clock speed would lead to better performance ?
Right now it seems just boosting, but maybe has no effect because not GPU limited ? I tried the Far cry 6 bench also at 'ultra', but didn't see much difference in fps between the min-max settings.
 

Attachments

  • Screenshot 2022-01-20 221207.png
    Screenshot 2022-01-20 221207.png
    29.7 KB · Views: 0
You can't just throw higher clocks at it. It will either become power limited or temperature limited. My guess is your power limited at 295 watts.
 
About 21000-22500 is what you will get on the 6900xt reference overclocked. Beyond 21k is luck of the draw.

I'm pretty sure stock was close to 20000 for me.
 
About 21000-22500 is what you will get on the 6900xt reference overclocked. Beyond 21k is luck of the draw.

I'm pretty sure stock was close to 20000 for me.
When did you bench your card? AMD drivers do age like fine wine. My score jumped 2,000 pts since last Feb.
 
When did you bench your card? AMD drivers do age like fine wine. My score jumped 2,000 pts since last Feb.
That's true (probably) only for the stock cards, if you overclock them you will get improvement from the driver if it fixes a problem with a game, which one is very, very rare and specific for a few games.
For me, this 'fine wine' is a myth. And this is very easy to be tested, I did it with my old RX 480, Vega 56 and my current 6700 XT.
You can get better FPS/scores if you repair/reinstall your Windows/VC++ libraries/broken video driver, but that's different.
 
That's true (probably) only for the stock cards, if you overclock them you will get improvement from the driver if it fixes a problem with a game

time to put down the bong, dude. it doesnt matter if the card is red or green...measurable performance uplift over time at driver level is extremely common with new architectures, either in average fps or frame pacing. you botching the measurement of a 5% reduction in peak frametime value (or whatever) doesnt make that iterative architectural optimization a myth.
 
time to put down the bong, dude. it doesnt matter if the card is red or green...measurable performance uplift over time at driver level is extremely common with new architectures, either in average fps or frame pacing. you botching the measurement of a 5% reduction in peak frametime value (or whatever) doesnt make that iterative architectural optimization a myth.
In the last several years, AMD never had a driver that is ready to use at the launch of the cards that they announced.
Four to six months after cards' launch they normally released fine working driver. If you count this as "improvement", ok.
After this time, if you overclock your video card to its limit and just change drivers you will see "fine wine" but other times :)

I prefer AMD cards but a few months after their launch.

The only exception to the moment, is with my current 6700 XT where the card gain 30-60MHz more overclocking room after a new driver. And this allows me to have a few FPS more in some benchmarks. But not for daily use, and only in specific benchmarks.

If you use your video card without overclocking, AMD has a lot of room to give you "fine wine" over years.
 
So went and tried my AMD 6900XT Reference with very little knowledge of Radeon GPUs(as I'm generally a Nvidia guy. VRAM: 2150, Min:2400/Max: 2800, Power: Max(15). Board was only pulling 293W. Next I'll learn to use MPT and see how much further I can take this card.

Coolant temp: 23C
Max GPU Temp: 36
https://www.3dmark.com/spy/25836539


20220125_004651.jpg
 
Last edited:
So went and tried my AMD 6900XT Reference with very little knowledge of Radeon GPUs(as I'm generally a Nvidia guy. VRAM: 2150, Min:2400/Max: 2800, Power: Max(15). Board was only pulling 293W. Next I'll learn to use MPT and see how much further I can take this card.

Coolant temp: 23C
Max GPU Temp: 36
Nice start. You're probably hitting power limit at 293W. What I would do is just up your PL in MPT to like 350. keep all the other settings, see if your score goes up. It should your Power Limit is now higher, and your clocks will be higher.
 
Back
Top