AMD Radeon VII Chilling & Undervolting @ [H]

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,534
AMD Radeon VII Chilling & Undervolting

We wanted to find out if there is anyway to improve Radeon VII power efficiency, GPU temperature, and fan noise without drastically changing the gameplay experience. Is Radeon Chill or GPU undervolting the answer? We run the Radeon VII through some real world gaming and show you exactly what Chill and Undervolting will do to, or for your gameplay.

*UPDATE 2/25*

We have made an update to the conclusion regarding GPU Temp / Fan Speed, please check it out right above The Bottom Line

https://www.hardocp.com/article/2019/02/24/amd_radeon_vii_chilling_undervolting/6
 
Last edited by a moderator:
Great article as always, and surprising results.

I always assumed undervolting would reduce performance, but I see that is not the case.

I've been having a lot of fun with overclocking the Radeon VII. Wattman has a ton of convenient options and it makes it easy to try things with minimal hassle.

As similar to the review, I also did not have much luck with Chill. I found it reduced performance too much and I'm usually on the edge (4K, etc.) and don't have the performance to spare.

One question: did you compare simply setting an fps cap in software (for example in RivaTuner) and would this result in similar savings?
 
If you can live with the higher temps, what a power savings you get by undervolting.
 
It looks to save 20-30 watts. Should be pretty close to the 2080 judging by Kingdom Come results in the original R7 review by [H].

Very professionally done article!
 
  • Like
Reactions: N4CR
like this
This article changed the game for me.

I was so focused on hitting 2000 MHz, I didn't really look into undervolting.

Tried some tests based on the article and I'm seeing about the same performance as 2000 on the core but with only 1880 MHz.

And with much better thermals, I don't even hear the fans spinning at all. Hopefully it will be more stable too (I was still having odd issues at 2000) but I've only been testing for about 1 hour so we'll see.

Thanks once again for rock solid reporting.
 
So it's quite like it's forefather in undervolting too, vegas seem to love undervolting on air or going hard with water and can have some sizeable gains for power envelope or maximum speed depending on how you are aiming. Fun to play with cards indeed, thanks for the neat side-review of this card.
 
Undervolting makes AMD cards perform faster? That is this heresy?
Lot of people write AMD off on the power front over a laughable amount of difference, whilst many users seem to get decent undervolting results but you don't hear of that from 'review' sites very often. AMD vcore bins are wider than that found in many Nvidia cards.
 
Lot of people write AMD off on the power front over a laughable amount of difference, whilst many users seem to get decent undervolting results but you don't hear of that from 'review' sites very often. AMD vcore bins are wider than that found in many Nvidia cards.

Can't ignore that the vcore has been set for stability across the range; one card might benefit significantly, while the next might crash with any droop.
 
Seems odd that so often the cards can be undervolted and perform better. Seems many can actually pull it off, which begs the question how many cards cant pull it off or is AMD just being overly cautious and running higher voltages to insure no issues.
 
Brent_Justice said:
Undervolting is hands down the "best" option to achieve all goals in our opinion. Undervolting reduces power without losing performance. It also improves performance and GPU clock speed. You are basically making your GPU more efficient by lowering the voltage, and there are little to no downsides

Nice analysis Brent , but can you explain something because otherwise i'm unable to understand :
since undervolting can result all these that you mentioned above (lesser power , same performance) ,then why exactly AMD doesn't sell this card with lesser voltage at default :eek:???
THERE MUST be some downside ,otherwise this means that AMD 's engineers sabotage their own GPU ,AND their customers as well , by making them pay more money at electrical bills for no reason whatsoever:wacky: !!!
(*i don't believe that AMD 's engineers are idiots , so there must be a reason for selling this card with these default settings. For example , when undervolting the components of the GPU , could this affect their longevity negative or positive? There must be some explanation which justifies their decision about the amount of voltage that they are using as default :confused:!!! )
 
Nice analysis Brent , but can you explain something because otherwise i'm unable to understand :
since undervolting can result all these that you mentioned above (lesser power , same performance) ,then why exactly AMD doesn't sell this card with lesser voltage at default :eek:???
THERE MUST be some downside ,otherwise this means that AMD 's engineers sabotage their own GPU ,AND their customers as well , by making them pay more money at electrical bills for no reason whatsoever:wacky: !!!
(*i don't believe that AMD 's engineers are idiots , so there must be a reason for selling this card with these default settings. For example , when undervolting the components of the GPU , could this affect their longevity negative or positive? There must be some explanation which justifies their decision about the amount of voltage that they are using as default :confused:!!! )

Not every card is going to undervolt as much and some might not even be able to at all. AMD has to set voltage at something that will work for every single card they produce not what might work on some.
 
What is the percentage of gpu's that can undervolt properly.

I heard it described as luck of the draw ,kinda like overclocking.

Is it guaranteed that every chip will be like the one in the review?
 
What is the percentage of gpu's that can undervolt properly.

I heard it described as luck of the draw ,kinda like overclocking.

I don't think anyone knows, I'm not even sure if AMD would be able to produce an entirely accurate number. If AMD was overestimating stock settings a bit then most cards should have some undervolting room, but its hard to say. You would need to get hundreds, if not thousands, of these cards and test them all to even begin to get a clear picture of how well they'll undervolt on average.
 
Lot of people write AMD off on the power front over a laughable amount of difference, whilst many users seem to get decent undervolting results but you don't hear of that from 'review' sites very often. AMD vcore bins are wider than that found in many Nvidia cards.


of the two brands AMD is definitely more conservative with their default voltages on gpu's. it's almost like they find 1 voltage that works across X number of GPU's and then set every gpu to that voltage to minimize product loss on potentially bad gpu's needing more voltage then all the others. either way i don't think it's a bad way to approach the issue but of course always gives something for people to complain about.

What is the percentage of gpu's that can undervolt properly.

I heard it described as luck of the draw ,kinda like overclocking.

Is it guaranteed that every chip will be like the one in the review?

it's still a lottery but most can be undervolted.. the key though is remembering that you can't lower the gpu voltage below the HBM voltage or at least that was the case with Vega so wouldn't be surprised if it was the same with radeon 7. average undervolt on air from what i've seen seems to be .98-.99v which isn't bad considering how much you could undervolt vega 56/64.
 
Last edited:
Capital article gents. Very interesting results indeed. I thought I was taking crazy pills seeing my temps go up a bit with an undervolt.

Now that we have seen the how low VII can limbo on voltage at stock speeds, will you be looking at how high it can clock stably? I've found with voltage a bit below default e.g. 1071mz, you can go well above 1900mhz core and 1150mhz on memory -- it seems to be thermals holding the card back...
 
Really good article Brent_Justice ..., thanks:)

............................................,THERE MUST be some downside ,..............

Funnily enough; when UV my Gigabyte RX580 8G @ ~ >-95mv i get not problems as such ( without specific testing; nothing comes to mind! ). Below -95mv i get screen transitions slowing ( by many seconds sometimes....; video-out block not recieving enough juice? ) and then eventually @ ~ or < -125mv, both screen transitions and games/desktop start to corrupt/crash; have been as low as -145mv but rekon i need MemV control to offset loss in voltage GPU-wide.
 
Last edited by a moderator:
We could speculate that AMD, uh, undervolts less than Nvidia as a cost saving measure; they could, after all, bin and test the GPUs and sell different SKUs like Nvidia's 'non-A' die that don't overclock as well. Binning each GPU, further than they do now if they're already doing it, would add to their cost per unit.
 
Seems odd that so often the cards can be undervolted and perform better. Seems many can actually pull it off, which begs the question how many cards cant pull it off or is AMD just being overly cautious and running higher voltages to insure no issues.

If you watched AdoredTV review it is just pressing a button on Wattman that allows this (2 clicks) and you are done you can do some more fine tuning yourself but that is it.
 
Good read.

it's still a lottery but most can be undervolted.. the key though is remembering that you can't lower the gpu voltage below the HBM voltage or at least that was the case with Vega so wouldn't be surprised if it was the same with radeon 7. average undervolt on air from what i've seen seems to be .98-.99v which isn't bad considering how much you could undervolt vega 56/64.

Vega gonna be a Vega, even at 7nm. My 64 undervolts to 0.98 and tries to push to almost 1600Mhz but can't take the heat due to this shitty blower cooler and holy fucking loud batman... I started looking at custom loop parts this weekend. We shall see.

It would be cool to see the [H] put that 7 underwater. I bet that joker will fly.
 
I haven't heard anything about long-term effects of undervolting. It is pretty much assumed that by overclocking, you could be shortening the life-span of your part, but what about undervolting? Anyone who has been doing that consistently over the course of a few years care to comment on stability?
 
just wondering why you didnt increase the power target? i know with vega56/64 you get the best results from undervolting and increasing the power target. did you test this and find that there was no performance gain from undervolting with an increased power target?
 
just wondering why you didnt increase the power target? i know with vega56/64 you get the best results from undervolting and increasing the power target. did you test this and find that there was no performance gain from undervolting with an increased power target?
Not tested in this article.
 
I don't think anyone knows, I'm not even sure if AMD would be able to produce an entirely accurate number. If AMD was overestimating stock settings a bit then most cards should have some undervolting room, but its hard to say. You would need to get hundreds, if not thousands, of these cards and test them all to even begin to get a clear picture of how well they'll undervolt on average.
AMD has that number, and you can bet has done much testing on it, which is why the card is set where it is. This is akin to overclocking in that some get golden chips, but you are only guaranteeed stock. I had a Fury X that wouldn’t overclock or under clock for much of anything. Errors and glitches either way. I had two Vega 56 launch reference cards that had almost a 100mhz stable undervolt testing delta between them. The engineers who design these things, surprise, know what they are doing. They aren’t purposely sabotaging anything Sith'ari
 
Last edited:
Great article and awesome feedback here. Going to try to undervolt my Sapphire Vega 64, already runs cool today and is a beast, but maybe she has more in her. So (anyone) use wattman or the Sapphire Trixx tool?
 
Interesting article. Would using a waterblock in a custom loop increase performance or just reduce temps for undervolting?
 
Interesting. My card runs at 1043-1045mv out of the box at stock settings. I am stable at 939mv but run at 950mv just to stave off any crashing.

In my testing I found that undervolting temps were higher because the fans ran around 700rpm lower than at the stock 1045mv.
 
Undervolt and overclock test!!! wonder if you can set it to say 1025 and get a few more MHz on the core etc.
 
Thanks for the excellent article [H], very interesting, with interesting results.

I am pretty sure that undervolting, since the temps are going up, is also increasing current draw through the GPU as the clock speed ramps up. Could we get AMD's feedback on the results?
Increased current drain puts more strain on the VRM's, and probably on the GPU as well. From a GN board component analysis (youtube), these cards have better than average VRM components. They might very well be able to easily handle that increased strain.

Would be a great follow up if there was some way to measure VRM temps in each of these same three scenarios of Default, Undervolted, and Chilled.

Thanks again!

Edit:
Did some quick current calculations.
Since the GPU voltage under default isn't exactly known, using the max default voltage of 1.084v. For undervolt, using .99v.

Using the power drain data on the Far Cry 5 page:
default settings: current drain estimates
Maximum: (280/1.084) 258.3A (holy shit, that's a lot of juice)
Average: (212/1.084) 195.6A
undervolted: current drain estimates
Maximum: (240/.99) 242.4A
Average: (186/.99) 187.9A

Probably not exactly right, as the voltage is likely moving around a bit in both scenarios. But as a rough estimate, it would appear that when undervolted, the current drain stays under the current drain estimated in the default settings. This is a good thing. But again, would love to hear and AMD engineers' comment on the current drain in both scenarios, and any side effects of undervolting.
 
Last edited:
Great article and awesome feedback here. Going to try to undervolt my Sapphire Vega 64, already runs cool today and is a beast, but maybe she has more in her. So (anyone) use wattman or the Sapphire Trixx tool?


I........, on my RX580; MUST only use MSI Afterburner for all OC related adventures...:-~ Any opening of Wattman inside General Settings causes instability to arise acrooss desktop use, that eventually forces a re-application of afterburner setings Grrrrrrrrrrrrrr : not happy:-(

Hmmmmmm..., have experienced the same thing happening when OC fails, forcing a Hard-Reset : ( CpU + Mem etc..., nothing to do with GPU! ); when stress/bench testing after reboot, not consistently achieveing same margin-of-error numbers; just enough outside of the standard distribution to be felt!

Followed by " Radeon settings has applied Defaults after a system error " ....... error............ F**K o** !!!!!!


Sorry Dudes n Gals......; been doing many hours of OC testing, right on the edges..., sick of these interupting messages from Radeon settings; they should turn OFF by Default :) problem solved.

Edit : spelling.
 
Undervolt and overclock test!!! wonder if you can set it to say 1025 and get a few more MHz on the core etc.


~ +6-8% range for me from stock, when undervolting; before the stock ( 0% ~ 1.156-1.161V ) must be used to stabilise further OC'ing. 1466/2100@+31mV - +50%PL - Highest sofar:)
RX 580 8G

Edit : clarity.
 
Last edited:
Has anyone tried compute heavy loads while undervolting? I'm curious to see if they require more voltage than gaming.
 
Back
Top