• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

RX 9000 series GPU Owners Club

and for me -80 is stable
You guys are lucky, I whish I could do -80mV too, it's -50mV here. I've been able to run -80 in many games, but many games also crash. I've settled for -50 which seems ok in all games.
 
You guys are lucky, I whish I could do -80mV too, it's -50mV here. I've been able to run -80 in many games, but many games also crash. I've settled for -50 which seems ok in all games.
Don't worry in windows I've settled with -50mV, cause i don't trust windows, and even if it was stable for gaming a few minutes, it's windows, and it's strangely worse drivers than in linux :D
Ah You could try to use moreclocktool or GPU Tweak III, with only driver from AMD, it could yield better results :). (but only if You wanna play around with drivers and stuff ^^)
And I've got windows Cause some of the older games have only support from modding community in windows, for example Ace Combat 7 is really easy to make it ultra wide in windows, but on linux I've crashed the game, and Nier Automata, or Replicant, has great tool to unlock framerates and other stuff that is working great with windows.
But Linux will come back, in new month, and new hardware (pc mobo and ram).
 
You guys are lucky, I whish I could do -80mV too, it's -50mV here. I've been able to run -80 in many games, but many games also crash. I've settled for -50 which seems ok in all games.
I'm kinda in the same boat now. I was able to run -80mV for a lot of games, including titles like God of War, Forza Horizon or Hogwart's Legacy, but Cyberpunk 2077 with some of its ray tracing forced me to dial it back to -50mV. So far, that's been the most stable undervolt for my scenario.

Current stable settings: -50mV, -50MHz, -4% PL — this combo has given me the best performance so far, and it’s been rock solid for hours now. For some reason, leaving PL or Clock limit at 0 gives me slightly worse results. Worth to mention.
Cheers :toast:

edit: In case you're wondering, this is on Windows 10 compilation 19045.5854 w/Adrenaline 25.5.1 drivers and AGESA 1.2.0.3b PatchC on the BIOS.
 
Last edited:
Current stable settings: -50mV, -50MHz, -4% PL — this combo has given me the best performance so far, and it’s been rock solid for hours now. For some reason, leaving PL or Clock limit at 0 gives me slightly worse results. Worth to mention.
I've had the same issue, stock power limit has strange spikes, I think highest was over 400W, and with power limit changed it has maybe over few watts, what is dialed.
 
For some reason, leaving PL or Clock limit at 0 gives me slightly worse results.
That's funny, because I have a similar issue with PL. I wanted to run PL @103%, it's 350W instead of 340W on my card, but end up crashing when I play for a long period. Even at -40mV. While PL @100%, -50mV, I'm fully stable. I now wonder if I could make -60mV happen with lower PL.

Also, I can't come to understand the difference between clock offset and the power limit. They both seem to do the something. If clock to let's say -200MHz, I don't really see a difference in clock, but TBP goes down, which is weird.

Edit: I'm doing the same test I was doing with older drivers and now changes to clock seems to do absolutely nothing. I'm trying -400MHz, 0 and +400MHz, and clock and TBP don't budge and inch.

Edit 2: using furmark to test that.
 
Last edited:
Also, I can't come to understand the difference between clock offset and the power limit. They both seem to do the something. If clock to let's say -200MHz, I don't really see a difference in clock, but TBP goes down, which is weird.
Max frequency offset is useless.
Voltage offset is all about stability and performance.
Power limit controls how much watts do you want to use and also performance.

My current 24/7 stable settings. Yes i can also set it at -80 easy but whats the point if it's not fully stable 24/7 ? All those fake benchmarks etc.... :D
Settings.png
 
Last edited:
My current 24/7 stable settings. Yes i can also set it at -80 easy but whats the point if it's not fully stable ? All those fake benchmarks etc....
I 200% agree with you on this. I like to get the best score I can with my card at least once just for fun, but when that's done, I want to find stable tunings asap.

Did anyone notice lower benchmark scores with latest AMD driver 25.5.1? I'm getting roughly 150-200 lower scores with same settings.
Before, I was able to score 7200+ score with out of the box settings. With the new driver, I get 7060-7080 using same settings.
It's the opposite here, I just made some Steel Nomad runs to know if I was in the same boat as you and it turns out I bested my previous score from previous drivers. 25.5.1 has been best driver so far for me. I have less annoying stutter in some games as well. I hope you find what's going on on your side.

New best score is now 7955, it was 7729 before.
110% PL (374W), -100mV, 2820MHz VRAM
 
Last edited:
Last edited:
That's funny, because I have a similar issue with PL. I wanted to run PL @103%, it's 350W instead of 340W on my card, but end up crashing when I play for a long period. Even at -40mV. While PL @100%, -50mV, I'm fully stable. I now wonder if I could make -60mV happen with lower PL.

Also, I can't come to understand the difference between clock offset and the power limit. They both seem to do the something. If clock to let's say -200MHz, I don't really see a difference in clock, but TBP goes down, which is weird.
I tried to set clock offset to -400 (maximum negative offset) and it decreseased boost limit by 400. GPU maximum boost was around 3060 MHz in that test.
It reached 294W TBP with 100% power limit and -70mV core offset. Before applying clock offset, max. reached boost clock (I have HwInfo sample rate set to 330 ms) was around 3170 MHz.
However, I did this test only twice, and as you might have read, I'm experiencing lower scores now for unknown reason, so it may not be accurate. Take it with grain of salt.

What seems to be 100% true is that clock offset applies to max boost limit which is shown in HwInfo.
My card has default limit set to 3450 MHz. Of course, even 110% power limit (363W) with -70mV on core voltage is not enough to reach this clock.
According HwInfo, there's a TDC limit (330A) which is hard coded and card won't go beyond that.
Anyway, when I apply clock offset, I see in HwInfo it immediately increases or decreases max. boost limit accordingly. That's in HwInfo.

Actually, raising power limit may cause instabilities when core is undervolted.
This is just for ilustration, I made up those values:
1747732402824.png

Undervolt will allow card to boost further (higher), because undervolt decreases power draw and remaining power draw headroom is used for auto OC (maximizing boost).
Higher power limit will allow card to boost further, because it increases power draw headroom and auto OC algorithm causes card to boost further.
Voltage-frequency curve is not linear. You require more and more voltage to reach stability at higher frequencies.
Thus, you may end up being unstable if you increase power limit and also apply undervolt, because it will try to boost higher and voltage might not be enough for that clock to be stable.

-90 offset, +10% PL
-90mV or -90MHz? :-)
 
Having good 3DMark results with the new software, slowly increasing memory frequency,.. no instability yet with GPU Tweak III.

 
Last edited:
Problem with legacy 3DMark benchmarks. Specifically 3DMark03, 3DMark05 and 3DMark06. Interestingly it does not affect 3DMark Vantage, 3DMark11 or the newest 3DMark suite first introduced in 2013 and continuously updated.

Ok so the problem with 03, 05 and 06 is that for some reason the GPU tests fps never exceeds monitor refreshrate (in my case 200Hz=200fps).
It should be near ~1150fps that is the max for these old, legacy benchmarks.

My GPU is AMD RX 9070 XT. I dont have any background programs running. FreeSync is disabled. V-Sync is forced Off manually. The description of the problem reads like something is forcing V-Sync on for these old benchmarks, but not for newer ones. I also disabled "Variable Refresh Rate" setting in Windows Display > Graphics > Advanced Graphics Settings. I will also add that on the same system i did not have this problem with Nvidia RTX 2080 Ti prior to replacing it with RX 9070 XT. I used DDU to fully remove Nvidia drivers before installing the AMD's card.

Can someone else please test this?
You can download and activate these legacy benchmarks from here: https://benchmarks.ul.com/legacy-benchmarks
03 is the smallest: https://benchmarks.ul.com/downloads/3dmark03.exe . Key is: N4LWE-T6HUX-8U4JB-28YLV
The problem should be visible within the first seconds of the first test. FPS should be ~1150 but is instead half of your monitors refreshrate or just below your max Hz number of your monitor (for this i had to create a custom profile with in the control panel and fiddle with settings not available in global default profile) to get it from ~99fps to ~199fps.

Naturally such limit would mean severely lower score. It's not a big problem, but it's just annoying.
 
Has anyone tested Cyberpunk 2077 with Patch 2.21, AMD FSR 3 Enabled.

Default:
Cyberpunk bench Default settings.png

Overclocked:
CP 2077 OC settings FSR 3.png
 
Last edited:
Problem with legacy 3DMark benchmarks. Specifically 3DMark03, 3DMark05 and 3DMark06. Interestingly it does not affect 3DMark Vantage, 3DMark11 or the newest 3DMark suite first introduced in 2013 and continuously updated.

Ok so the problem with 03, 05 and 06 is that for some reason the GPU tests fps never exceeds monitor refreshrate (in my case 200Hz=200fps).
It should be near ~1150fps that is the max for these old, legacy benchmarks.

My GPU is AMD RX 9070 XT. I dont have any background programs running. FreeSync is disabled. V-Sync is forced Off manually. The description of the problem reads like something is forcing V-Sync on for these old benchmarks, but not for newer ones. I also disabled "Variable Refresh Rate" setting in Windows Display > Graphics > Advanced Graphics Settings. I will also add that on the same system i did not have this problem with Nvidia RTX 2080 Ti prior to replacing it with RX 9070 XT. I used DDU to fully remove Nvidia drivers before installing the AMD's card.

Can someone else please test this?
You can download and activate these legacy benchmarks from here: https://benchmarks.ul.com/legacy-benchmarks
03 is the smallest: https://benchmarks.ul.com/downloads/3dmark03.exe . Key is: N4LWE-T6HUX-8U4JB-28YLV
The problem should be visible within the first seconds of the first test. FPS should be ~1150 but is instead half of your monitors refreshrate or just below your max Hz number of your monitor (for this i had to create a custom profile with in the control panel and fiddle with settings not available in global default profile) to get it from ~99fps to ~199fps.

Naturally such limit would mean severely lower score. It's not a big problem, but it's just annoying.
Mate, you wanna kill my card :laugh:
The FPS jump to near 9900+ sometimes :D

1747765756222.png

**************
OK, this time I up some settings in the benchmark :)
But, yeah, no problem with high FPS here.

1747767857017.png


****
3dMark05 with online result

1747768871955.png



****
And the last one, this time the online result also shows GPU results :)
But on this resolution, GPU is running at 35-45% at times 60%
And the main menu doesn't shrink because of full-screen mode.

1747769698142.png


 
Last edited:
I'm playing really demanding title :) unreal gold :), it's fully stable and runs butter smooth with this gpu :D it was great investment :D

unreal gold.gif
 
Mate, you wanna kill my card :laugh:
Interesting results, thanks for sharing! These benchmarks use dx 7 to dx 9.0c, so everything should be CPU bound, especially in the low default resolution.

I'll test my stock MBA 7900xtx later to see how it does against the 9070xt here :p

I'm playing really demanding title :) unreal gold
Love this game! I play old(er) titles on my main rig 99% of the time. Currently on a UE3 game from 2009. I'm playing it in 4K maxed out + ini tweaks for quality, with forced transparency MSAA 8xEQ (16x effective).
The card is capped to its minimum 600 MHz GPU clock at 700 mV in Adrenalin. Yet I'm seeing below 75% utilization :toast:
 
Last edited:
I tried to set clock offset to -400 (maximum negative offset) and it decreseased boost limit by 400. GPU maximum boost was around 3060 MHz in that test.
It reached 294W TBP with 100% power limit and -70mV core offset. Before applying clock offset, max. reached boost clock (I have HwInfo sample rate set to 330 ms) was around 3170 MHz.
However, I did this test only twice, and as you might have read, I'm experiencing lower scores now for unknown reason, so it may not be accurate. Take it with grain of salt.

What seems to be 100% true is that clock offset applies to max boost limit which is shown in HwInfo.
My card has default limit set to 3450 MHz. Of course, even 110% power limit (363W) with -70mV on core voltage is not enough to reach this clock.
According HwInfo, there's a TDC limit (330A) which is hard coded and card won't go beyond that.
Anyway, when I apply clock offset, I see in HwInfo it immediately increases or decreases max. boost limit accordingly. That's in HwInfo.

Actually, raising power limit may cause instabilities when core is undervolted.
This is just for ilustration, I made up those values:
View attachment 400378
Undervolt will allow card to boost further (higher), because undervolt decreases power draw and remaining power draw headroom is used for auto OC (maximizing boost).
Higher power limit will allow card to boost further, because it increases power draw headroom and auto OC algorithm causes card to boost further.
Voltage-frequency curve is not linear. You require more and more voltage to reach stability at higher frequencies.
Thus, you may end up being unstable if you increase power limit and also apply undervolt, because it will try to boost higher and voltage might not be enough for that clock to be stable.


-90mV or -90MHz? :-)
Thanks for sharing! I now understand why I didn't saw a variation in clock while runing furmark. My limit is also 3450MHz. -400 bring the limit to 3050MHz and clock speed is around 2860MHz while runing furmark. Not even close to the limit.

Now that I understand what clock offset really does, I have it set to -250. I don't see the point of going over 3200MHz while undervolting. Having the limit set to 3450MHz is just asking for trouble in my opinion.
 
Mate, you wanna kill my card :laugh:
The FPS jump to near 9900+ sometimes :D

View attachment 400468
**************
OK, this time I up some settings in the benchmark :)
But, yeah, no problem with high FPS here.

View attachment 400475

****
3dMark05 with online result

View attachment 400488


****
And the last one, this time the online result also shows GPU results :)
But on this resolution, GPU is running at 35-45% at times 60%
And the main menu doesn't shrink because of full-screen mode.

View attachment 400489

Much appreciated mate. Now i know it's not a universal thing and the problem is somewhere in my PC. Hopefully it's software and i can get it fixed by changing some setting i have overlooked. If it's not too much trouble can you provide screenshots of the following settings:

Windows > System > Display > Graphics. Make sure advanced graphics settings options are also visible. Apps list it not necessary.
AMD Adrenalin Control Panel > Gaming > Graphics. I want to see what profile is in use and what Graphics Options list shows.
AMD Adrenalin Control Panel > Cogwheel (Settings) > Display. Make sure Display Specs and Overrides options are also visible.
 
Much appreciated mate. Now i know it's not a universal thing and the problem is somewhere in my PC. Hopefully it's software and i can get it fixed by changing some setting i have overlooked. If it's not too much trouble can you provide screenshots of the following settings:

Windows > System > Display > Graphics. Make sure advanced graphics settings options are also visible. Apps list it not necessary.
AMD Adrenalin Control Panel > Gaming > Graphics. I want to see what profile is in use and what Graphics Options list shows.
AMD Adrenalin Control Panel > Cogwheel (Settings) > Display. Make sure Display Specs and Overrides options are also visible.
I may later today, but you can see if you have the Triple Buffering checkbox turned on in AMD Adrenalin Control Panel > Gaming > Graphics, some games may interpret it as "Vsync".
This has probably been a bug for a while.
 
I'm playing really demanding title :) unreal gold :), it's fully stable and runs butter smooth with this gpu :D it was great investment :D
Don't forget to ramp up RT settings to get that RX 9070 XT properly utilized! :D

Thanks for sharing! I now understand why I didn't saw a variation in clock while runing furmark. My limit is also 3450MHz. -400 bring the limit to 3050MHz and clock speed is around 2860MHz while runing furmark. Not even close to the limit.

Now that I understand what clock offset really does, I have it set to -250. I don't see the point of going over 3200MHz while undervolting. Having the limit set to 3450MHz is just asking for trouble in my opinion.
Clock offset can save your ass when your undervolt is pretty on the edge of stability. Let's say you are stable with -100mV until 3100 MHz, but anything past 3100 MHz causes crash.
If this clock speed is enough for you and you prefer efficiency over max. performance, you can limit core clock by this option to always stay below critical point and thus achieve stability.
Maybe after 3100 it would be stable with -80mV, but it would consume 30W more, which renders is irrational. (I made up values.)
 
Mate, you wanna kill my card :laugh:
The FPS jump to near 9900+ sometimes :D

**************
OK, this time I up some settings in the benchmark :)
But, yeah, no problem with high FPS here.


****
3dMark05 with online result



****
And the last one, this time the online result also shows GPU results :)
But on this resolution, GPU is running at 35-45% at times 60%
And the main menu doesn't shrink because of full-screen mode.



LOL, frime time <0ms. How's that possible?
 
Sure, but interpreting something that is 0.00000000001ms as something <0 is like really badly incorrect.
If there were 0 ms, I'd say it's fine. But -1ms? Frame time can't be of negative value. Doesn't matter, but nice bug.
 
Back
Top