[SOLVED] Need some help with undervolting my RX 580

Karizma

Distinguished
Mar 28, 2012
16
0
18,510
Hello I have a Rx 580 armor that has been running hot and loud and I wanted to try to undervolt it using wattman.

So far I set Power limit to 50 which is max and I brought down the voltage for State 7 to 1005 and I don't seem to be having any crashing or issues in game with this and the temps have gone down some and so have the fan noise.

So now my question is state 3,4,5, and 6 all have higher values than state 7 so do I need to lower those and if so what should I put them to. Now with those things done is there anything else I need to change or would that be it, this is the first time I've ever messed with the setting of a gpu before.
 
Solution

Hit the white toggle button next to "Frequency %"

TJ is right. If you DON'T enable Chill/FRTC/Power Efficiency, your GPU will mostly either run at State 7 or idle. You may get some sporadic bounces into the lower states, but it shouldn't be frequent/prolonged.

Also, forgot to add, your VRAM voltage sets the lowest the core voltage will go (except for idle state). Minimum VRAM voltage is a little harder to test for, basically you need to give the core "plenty" of voltage so you take it out of the stability equation and then lower the VRAM voltage until you reach instability. I haven't done extensive testing to find...
1005mV? Wow, at what frequency? 1366MHz? That's impressive.

Also, what program are you stress testing with. I initially stress tested with FurMark for years, but recently found that F@H (of all things) is much more thorough at exposing instabilities (good for finding faulty RAM much quicker than MemTest or Prime95 also...a story for another time). Typically F@H for 2 hours is enough to prove very good stability. Once I got my settings dialed in from that, I tested for a full 24hrs and nothing changed.

Power Limit increases shouldn't be necessary when you're undervolting. Basically all that setting is doing is letting the GPU auto-overvolt itself. When you go to manual voltages, that ability gets taken away so the card will either run stably (with sufficient manual voltage setting) or it will crash (insufficient manual voltage setting)

If you give me the frequency of each state, I can ballpark you a voltage.

What you want to do is test out a voltage at the State 5 frequency. Then you can begin to generate a graph (ie Excel) of your freq/voltage curve so you can interpolate which saves you time of testing every single state. GloFo's 14nm process inflects at around 950mV. So expect any states above that point to have a steeper voltage curve than states below it.

Some other things to compound benefits:
If you have a FreeSync monitor, enable "Chill"
If you don't have a FreeSync monitor, enable "Frame Rate Target Control"
I've played around with the "Power Efficiency" toggle, and while it didn't adversely affect frame rate consistency much, it seems to randomly disable itself on my machine, so I just gave up having to check it was on every time before I opened a game. The power savings wasn't as extreme as Chill/FRTC.
 


Maybe I was doing it wrong because I was just going into some high settings games and seeing if it would cause artifacting or crashing or anything, I didn't actually use a stress test program. Should I do that or is game testing enough?

I'm not sure how to see individual frequency only All I could see is a general one that is at Zero, This is my wattman page as of now.
 
I agree that 1005 mV is quite low if you haven't also underclocked. But if you're not experiencing any crashes or issues in games, your undervolt is probably stable enough for your purposes. If you experience any issues in the future, you can raise voltage a bit if need be.

There shouldn't be any issues with the lower p states having higher voltages. I've found that my card rarely enters the intermediate states anyway, it's either at the lowest state while doing regular desktop stuff and at the highest while gaming.

Basically, you're probably fine leaving everything as you have it if everything seems to be working fine.
 

Hit the white toggle button next to "Frequency %"

TJ is right. If you DON'T enable Chill/FRTC/Power Efficiency, your GPU will mostly either run at State 7 or idle. You may get some sporadic bounces into the lower states, but it shouldn't be frequent/prolonged.

Also, forgot to add, your VRAM voltage sets the lowest the core voltage will go (except for idle state). Minimum VRAM voltage is a little harder to test for, basically you need to give the core "plenty" of voltage so you take it out of the stability equation and then lower the VRAM voltage until you reach instability. I haven't done extensive testing to find absolute minimum, but I run 2000MHz/~900mV for the VRAM. Therefore, any performance states that I've set a core voltage <900mV will just get 900mV anyway. Regardless of manual core voltage settings.
 
Solution
The premise of FRTC is that not all games require the full potential of your GPU (lightweight games such as LoL, Overwatch, Warframe, Doom, etc etc). Therefore, allowing the GPU to intelligently clock itself in such a manner that it is only producing frames that your monitor can actually display (ie 60Hz = 60fps) has a large potential to save power in games where your GPU can produce, say, 100-200FPS at it's highest clockspeed. For fixed refresh monitors, you have to be a little careful with this setting because of frame rate variance (perhaps you set it at 75FPS so you're always in 60Hz VSync), but you get the drift.

Chill takes that same concept and adds the ability for the frame rate to be reduced in scenes/instances where there's little/no movement on screen/in game. For these situations, perhaps 40FPS is sufficient to maintain acceptable visuals. Here you get a maximum AND a minimum frame rate slider to move around at your discretion.

I'm not entirely sure what set of rules "Power Efficiency" is operating on. Sorry. I imagine it's more of a feature that's used in "auto" voltage operation. (maybe that's why it keeps getting disabled!! I'll have to look into this)

**Word of caution - for whatever reason, my PC doesn't always want to correctly apply my manual undervolt profile on cold boot (make sure you save your undervolt profile so you can simply "load profile"). Oftentimes, I have to restart (not shut down and reboot) and then it applies properly. This...bug applies to some but not all of other PCs I've worked on.
 
Try:
920MHz/850mV
1165MHz/915mV
1240MHz/945mV
1280MHz/980mV
1325MHz/1020mV (a bit lower than I'd expect (~1040mV), but based on your impressive seemingly stable results....)
1365MHz/1070mV (again a bit lower than I'd expect (~1110mV))


VRAM = 900mV
 
I really don't know if all that's necessary. The OP said he wanted to try and make his GPU run cooler and quieter, and his current undervolt for state 7 should achieve that. If you're running at lower p states heat shouldn't be a problem to begin with, and undervolting the VRAM wouldn't make much difference for core temps or noise.

If the OP wants optimal efficiency and power draw across the board though, that's another story.
 
It wasn't my intent to force the OP deeper into the rabbit hole that he/she wanted, but why not lay out the whole playing field and let them choose how far to run?
The OP came to the table implying(?) that they already had the GPU undervolted at its highest boost frequency, but wanted to know more about undervolting the other P states.

Also, as I mentioned, using FRTC/Chill is additional power savings above and beyond undervolting, but if you're letting the GPU auto-set the lower P states to the same or more voltage than State 7, it defeats the purpose. You could generate a freq/voltage curve for all the P states once and let FRTC/Chill do the rest, or you can achieve the same result by determining what GPU frequency is required to produce xx FPS in each individual game and manually tune only State 7 frequency/voltage for that game profile. But that's a real pain in the...