[SOLVED] Help with undervolting Rx570

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Recently attempted to undervolt an rx570 in a system as its temperatures were a little on the high side. Started by lowering the state 7 core voltage by around 30mV each time and then ran heaven benchmark to see the results with adrenaline metrics overlay running to show statistics. The confusing thing was at as low as 1075mV (Default is 1150) I was still getting higher wattage used by the GPU (around 115 W as opposed to around 100-105 at default, power limit set the same for both). While the default kept the core clock constant the undervolt actually caused it to throttle down slightly, before heat even became an issue. Am I just missunderstanding this process? Advice would be really appreciated, cheers
 
Last edited:
Solution
As an FYI, here's some points along the curve that I've generally found stable for most cards.

1000MHz = 890mV
1100MHz = 910mV
1200MHz = 940mV (this 940mV is typically an inflection point on the curve, it gets steeper above this value. Notice that values below this are ~25mV/100MHz and values above are 100mV/100MHz)
1300MHz = 1030mV
1400MHz = 1150mV

If the stock frequency was 1244MHz, it would seem to me that auto voltage would probably feed around 1075mV (even though it may have been higher when you unlocked it in the WattMan window). In that case, the GPU may have been getting ~1075mV on Auto voltage anyway. Did you see 1150mV in GPUz for a significant period of time (or was it just a brief spike)?

My personal MSI Gaming X RX480...
Sounds like you're lowering the voltage, but leaving the GPU on the default frequency/voltage curve, so if it wants 1150mV for 1300MHz, and 1075mV for 1200MHz, it's just setting the frequency to whatever voltage that corresponds to on the default chart. This is easy to override. Not sure why wattage is going up though...

What software are you using? WattMan or Afterburner?

What software are you using to monitor the metrics on the GPU (actual voltage, temps, wattage, etc)

Here's an undervolting guide I put together for WattMan
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Thanks for the reply, I am using Radeon adrenaline 2020 that comes with the drivers for both monitoring and changing settings I assume that is the newer equivalent of wattman? Would it be better to use something like hwmonitor for monitoring?
I'll be honest im a bit confused by your response, I was under the impression that state 7 was the level at which the card would run under max load and therefore by begining with lowering voltages there I would be able to test their stability before moving to lower states or am I missunderstanding how this works? And yes my main confusion is to be honest why wattage is increased, surely it should be using less power.

Sounds like you're lowering the voltage, but leaving the GPU on the default frequency/voltage curve, so if it wants 1150mV for 1300MHz, and 1075mV for 1200MHz, it's just setting the frequency to whatever voltage that corresponds to on the default chart. This is easy to override. Not sure why wattage is going up though...

What software are you using? WattMan or Afterburner?

What software are you using to monitor the metrics on the GPU (actual voltage, temps, wattage, etc)

Here's an undervolting guide I put together for WattMan
 
Yes, Adrenalin 2020 is the overall name of the AMD driver software. WattMan is a utility included in that software.

I just use GPUz. It's easy. WattMan has a graph of all the pertinent info also, I just like the way it's presented on GPUz better.

Have you read through the guide I linked?

Are you set on % or Dynamic Frequency?

Did you make sure that none of the other lower states' voltages were exceeding the voltage you set for State 7? What is the VRAM voltage set at?
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Ah my mistake, and yes I have just briefly read it through. It looks like I have skipped the memory voltage aspect of the undervolt. Could this be acting as a lower limit for my GPU core voltage as you discuss in the guide? This still doesn't explain the higher wattage though

Yes, Adrenalin 2020 is the overall name of the AMD driver software. WattMan is a utility included in that software.

I just use GPUz. It's easy.

Have you read through the guide I linked.
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Just ran a test with all states lower than 1075 and I'm having similar results, core clock jumps around in th low 1200s utilisation also goes from 90-100. 112w power usage. Like I say stock is about 105w. Vram voltage is stock at 900mV. Not sure what you mean by % or dynamic frequency, I am adjusting individual states voltages and frequencies on the graph. The only voltage number I can see on GPUz is vddc which was going from 1.03 to 1.05 throughout

I added some comments to my previous post while you were responding. Sorry.
 
Last edited:
What software are you using for wattage reporting? WattMan?

There's no way the GPU is using less power at stock 1175mV than at your manual 1075mV. That might just be a reporting error. Whats the fan rpm and temps for stock and manual? That should be an approximate estimate of power draw.

What software are you using to stress test?

Is 1200MHz the frequency setting you have in State 7, or? It's unclear what significance this is. What's the stock/auto voltage frequency?

% or Dynamic Frequency are the two options on the far left side of the frequency row (a toggle switch). One shows a % and the other shows 7 frequencies in MHz. I assume you're on Frequency since you're talking about State 7.
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Fan and temps are pretty much exactly the same as at stock. Both GPUz and Wattman report these wattages. I'm using heaven benchmark to test.
The frequency set in state 7 is 1244
Its default voltage is 1150mV

What software are you using for wattage reporting? WattMan?

There's no way the GPU is using less power at stock 1175mV than at your manual 1075mV. That might just be a reporting error. Whats the fan rpm and temps for stock and manual? That should be an approximate estimate of power draw.

What software are you using to stress test?

Is 1200MHz the frequency setting you have in State 7, or?

% or Dynamic Frequency are the two options on the far left side of the frequency row (a toggle switch). One shows a % and the other shows 7 frequencies in MHz. I assume you're on Frequency since you're talking about State 7.
 
Last edited:
As an FYI, here's some points along the curve that I've generally found stable for most cards.

1000MHz = 890mV
1100MHz = 910mV
1200MHz = 940mV (this 940mV is typically an inflection point on the curve, it gets steeper above this value. Notice that values below this are ~25mV/100MHz and values above are 100mV/100MHz)
1300MHz = 1030mV
1400MHz = 1150mV

If the stock frequency was 1244MHz, it would seem to me that auto voltage would probably feed around 1075mV (even though it may have been higher when you unlocked it in the WattMan window). In that case, the GPU may have been getting ~1075mV on Auto voltage anyway. Did you see 1150mV in GPUz for a significant period of time (or was it just a brief spike)?

My personal MSI Gaming X RX480 8GB supplies ~1075-1125mV @ 1305MHz when left on Auto voltage.
 
Last edited:
Solution

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Ah ok this makes sense, i never tested stock with wattman, but with the undervolt as mentioned the voltages shown were actually even lower than 1075 (around 1030-1055). So the problem here is that I essentially haven't got to the point at which I'm undervolting compared to auto and need to take it even lower do you think?

As an FYI, here's some points along the curve that I've generally found stable for most cards.

1000MHz = 890mV
1100MHz = 910mV
1200MHz = 940mV (this 940mV is typically an inflection point on the curve, it gets steeper above this value. Notice that values below this are ~25mV/100MHz and values above are 100mV/100MHz)
1300MHz = 1030mV
1400MHz = 1150mV

If the stock frequency was 1244MHz, it would seem to me that auto voltage would probably feed around 1075mV (even though it may have been higher when you unlocked it in the WattMan window). In that case, the GPU may have been getting ~1075mV on Auto voltage anyway. Did you see 1150mV in GPUz for a significant period of time (or was it just a brief spike)?
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Ok great news, to be honest I had no idea auto voltage altered from those set states and that was its own independent setting. Thanks for your help, will report back and let you know how the undervolting goes, might even try a slight overclock to around 1350ish at the voltage im currently running, as a slightly more aggressive fan curve seems to be able to keep it under 70c at this voltage. I certainly have a few things to try out

 
Just an FYI, I re-applied thermal paste to my RX480 a year ago and temps went from 85C @ 2400rpm to 73C @ 900rpm. Using the same undervolt settings. Even when my card was brand new, it wasn't fantastic (maybe halfway between those), but temps had been slowly degrading over the 9-12 months prior to repasting.
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Yeah I'm thinking I might do this. Was trying to avoid it as the Gigabyte 4g model I have has four really awkward screws on the shroud that you have to access through little gaps, making it easy to loose the screws etc, however I think it is probably a good idea. Thanks again!

Just an FYI, I re-applied thermal paste to my RX480 a year ago and temps went from 85C @ 2400rpm to 73C @ 900rpm. Using the same undervolt settings.
 
Those might just be for removing the shroud from the heatsink fins. Usually simply removing the heatsink with shroud attached is all you need. Those screws are always on the back side of the card from what I've seen.
Also, not all the screws on the back of a GPU are necessarily for removing the heatsink. If your card has a backplate, there will likely be screws solely dedicated to holding the backplate to the PCB. No need to remove those when you're trying to take off the heatsink for a repaste.

Start with the 4 screws on the back of the card around the GPU chip itself. Some cards only have those 4 screws holding the heatsink down (mine is one of them).
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Oh really ok, I'll try that thanks

Those might just be for removing the shroud from the heatsink fins. Usually simply removing the heatsink with shroud attached is all you need. Those screws are always on the back side of the card from what I've seen.
Also, not all the screws on the back of a GPU are necessarily for removing the heatsink. If your card has a backplate, there will likely be screws solely dedicated to holding the backplate to the PCB. No need to remove those when you're trying to take off the heatsink for a repaste.

Start with the 4 screws on the back of the card around the GPU chip itself. Some cards only have those 4 screws holding the heatsink down (mine is one of them).
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
So it turned out it was just the six screws at the back of the card I needed to take off like you said, was pretty easy! Thermals are a bit better with some mx4 applied, coupled with a 950mV undervolt running at ~ 85-95w everything looks good. I could get it a bit lower I think, but would rather be secure in its stability. Thinking I'll put a fan on the front as an air intake as the case only has an exhaust and is still a bit of a hot box, plus I have a random 120mm fan lying around. With that done the system should be ready to go. Thanks for you help

Those might just be for removing the shroud from the heatsink fins. Usually simply removing the heatsink with shroud attached is all you need. Those screws are always on the back side of the card from what I've seen.
Also, not all the screws on the back of a GPU are necessarily for removing the heatsink. If your card has a backplate, there will likely be screws solely dedicated to holding the backplate to the PCB. No need to remove those when you're trying to take off the heatsink for a repaste.

Start with the 4 screws on the back of the card around the GPU chip itself. Some cards only have those 4 screws holding the heatsink down (mine is one of them).
 
Yes. My general rule of thumb for cooling:
  1. A <$500 system with no graphics card (IGP only) can get by with a single case fan (generally the rear exhaust fan that comes standard in most cases)
  2. A system with a graphics card not needing external power (75W TDP, GTX1050Ti, etc) gets 1 intake + 1 exhaust. Your RX570 is close to this cutoff, so maybe your fans just spin slightly faster (say 1200-1400rpm)
  3. A system with a graphics card <$350-ish gets 2 intake + 1 exhaust (fan speed can vary here, lower TDP cards like your RX570 would have slow spinning fans, an RX5700 would have larger or slightly faster spinning fans)
  4. Above that, more fans needed.
 

Robbom

Reputable
Mar 13, 2016
70
1
4,545
Ok thanks that's really helpful info, like you say hopefully one fan will be enough or it should at least improve the situation a bit. I was thinking mount the front intake fan low to push cold air over the GPU? The ryzen 1600 in the system is rarely fully pegged unlike the GPU and it's right by the exhaust anyway, plus this would pull in cold air lower down and expell hot air higher up.

Yes. My general rule of thumb for cooling:
  1. A <$500 system with no graphics card (IGP only) can get by with a single case fan (generally the rear exhaust fan that comes standard in most cases)
  2. A system with a graphics card not needing external power (75W TDP, GTX1050Ti, etc) gets 1 intake + 1 exhaust. Your RX570 is close to this cutoff, so maybe your fans just spin slightly faster (say 1200-1400rpm)
  3. A system with a graphics card <$350-ish gets 2 intake + 1 exhaust (fan speed can vary here, lower TDP cards like your RX570 would have slow spinning fans, an RX5700 would have larger or slightly faster spinning fans)
  4. Above that, more fans needed.
 
Aug 1, 2020
5
0
10
MHz determines FPS, not voltage.
im understand that but, someone on yt says it may be: (its only beacouse this im asking)
View: https://www.youtube.com/watch?v=k53gmCxdj0o



leanhk

Undervolt too much will cause stutter / frame drops, shouldn't just undervolt until the game crash. You should also monitor the 1% low FPS, if 1% low FPS starts to drop, which means you are undervolting too much. Sweet spot is about 50-60mV