Undervolting an RTX 2080.. ?

Orpheus

Bronze Level Poster
Hey guys,

I've been playing Dishonored 2 at max settings lately, but my RTX 2080 seems to be a bit hot for my liking.
It's not a super demanding game and the GPU still reaches temps between 86°C and 89°C maximum (with rare peaks at 90°C).
I would like to know if temps like that are actually ok in the long run, or should I undervolt it a little bit using MSI Afterburner ? Could my warranty be void by doing so ?

In the screenshot you can see I've been hitting the temperature limit, and it seems to be at 86°C. I also hit the power limit sometimes.
Keep in mind that those temps aren't reached over time but rather in the first few seconds following the launching of the game.

I've also been reducing the CPU cores to their base clock (3.6 GHz) so it stays pretty cool during gaming (64°C).

What would you recommend ?

Thanks a lot !
 

Attachments

  • GPU Temp.png
    GPU Temp.png
    866.8 KB · Views: 500

FerrariVie

Super Star
Keep in mind that power limit is totally normal in laptops at stock, so if you're not getting those, it's either because your game is not demanding and is not using your GPU fully or because there is another type of limit (in your case thermal) that is limiting the GPU. I would recommend you also add the GPU power usage on the RTSS overlay, so you can have an idea about that while gaming.

I personally prefer losing around 10-15 fps to not have power limit throttling, as in my opinion is the best way to efficiently use the GPU: lower power, lower temperature and a bit lower performance, but still good enough to play my games at ultra settings with 80+ fps and keep the fans quieter. To achieve this, I am indeed using MSI afterburner to undervolt the GPU. I have a few different profiles that I use with different levels of undervolt, depending on how demanding the game is. You need to find your sweet spot. PS: That won't void your warranty, as it doesn't damage your GPU. It's the opposite, actually... if you run it cooler, you're very likely prolonging its lifetime. Overvolt, on the other side, can cause damage and will probably void your warranty, so avoid doing that (I don't think you need it on a 2080 :ROFLMAO: )

Regarding your thermals, do you mind posting which laptop do you have? Full specs, if possible. It will help me and others here to understand what the issue might be.
 

Orpheus

Bronze Level Poster
Hey there !
Thanks for the reply.

I couldn't find the GPU Power Usage in the Afterburner listing, does it have a specific name ?

As for the specs, I'm using an Octane V (based on the Clevo P751TM1-G chassis) 15.6" 144Hz, i9 9900K, RTX 2080, 32GB of 2666MHz RAM.

I've never done a GPU undervolt before so I would love to hear some of your thoughts on how to do it properly.

Thanks !
 

Attachments

  • Specs.png
    Specs.png
    47.1 KB · Views: 364
  • XTU specs gaming.png
    XTU specs gaming.png
    29.3 KB · Views: 362
  • XTU specs normal.png
    XTU specs normal.png
    37 KB · Views: 370

FerrariVie

Super Star
Wow, that's a lot of heat to dissipate :LOL: . You have such a powerful laptop, so I don't think there's nothing wrong with it. Even though a repaste might help, tweaking voltage and clocks will be the way to go to tame your beast.

I learned the basics of GPU undervolting by watching BOAT's videos. Check out this one:


You don't need to follow his advice completely, but it's a good start. Oh, and definitely do not change the dots one by one, do this instead:
After you chose the dot that you want to use as the max, hold shift and select all the consecutive ones with LMB. Then press Shift + Enter(2x), it will then make the line perfectly straight.
 
Last edited:

FerrariVie

Super Star
Below are a few of my profiles (2070), just as an example:

Stock (AAA):
1602091403517.png

Higher voltage possible to not have power limit throttling:
1602091441829.png

Old games:
1602091334037.png

And by the way, the power usage of the GPU is simply called "Power" on Afterburner. I know it could be different between mobos, though. My old Dell 7559 did not have this option.

1602091683699.png
 

Orpheus

Bronze Level Poster
Thanks a ton man, this will definitely help !

Funny thing since I started this thread, now the GPU won't go over 80°C even though I didn't change anything since last time. I'm pretty confused about that but it's nice. So I've been running HWinfo just like Bob did in his tutorial but I'm a bit confused about the value (or values) I should take into consideration to make my custom curve.

At which point should I make the curve itself rise to hit that straight line ? Should I cap the voltage at 1000 since my voltage under load is 1.069V ?

By the way, my RTX 2080 clocks waaaay way over its boost clock of 1590MHz, I'm sitting at 1890MHz with power/voltage limit.
 

Attachments

  • GPU hwinfo.png
    GPU hwinfo.png
    45.1 KB · Views: 553
  • Stock curve RTX 2080.png
    Stock curve RTX 2080.png
    17.5 KB · Views: 391

FerrariVie

Super Star
Thanks a ton man, this will definitely help !

Funny thing since I started this thread, now the GPU won't go over 80°C even though I didn't change anything since last time. I'm pretty confused about that but it's nice.
No problem at all!

It's probably performance season coming (winter :p ). Ambient temperatures will have influence over the temps on your laptop, as well as the time you've been stressing it (gaming). Like if you only play for an hour or two, the temps will be lower than when you play for 6 or 8 straight hours.

So I've been running HWinfo just like Bob did in his tutorial but I'm a bit confused about the value (or values) I should take into consideration to make my custom curve.

At which point should I make the curve itself rise to hit that straight line ? Should I cap the voltage at 1000 since my voltage under load is 1.069V ?
You could start by the max boost clock specified by your GPU, but I found that to lose a lot of performance, even though the drops in temperature for the GPU were huge, always below 68° in my case, no matter what I throw at it. So what I did was try to find the sweet spot, the highest frequency and voltage I could get before power limit throttling started kicking in. To check that, just add the power limit flag to MSI/RTSS, so while gaming you will see at which clock the power limit starts. That way I could squeeze a bit more performance out of my GPU, while still keeping temps below stock.

After you found that sweet spot, the tweaking starts. You can raise it a bit more, because even though you will have a power limitation, it won't be much. So raising the voltage a bit will give you higher clocks and better performance, but a bit more heat as well. So it's up to you to decide what temperatures you want the GPU to work on. I like to have mine below 75° while having fans at 60%, but that is totally a personal preference for heat and noise. Another option would be to raise your custom fan curve on control/gaming centre to have the fans ramp up to max a lot quicker and keep you GPU at full power and voltage at all times while still keep the temps low, but that's not what I want.

So let's assume that your sweet spot will be at around 1775Mhz and 950mv. In this case, your curve should look like this (in red):
1602145794381.png

By the way, my RTX 2080 clocks waaaay way over its boost clock of 1590MHz, I'm sitting at 1890MHz with power/voltage limit.
That is normal and happens to all GPUs, it's kind of an extra boost. My 2070 refresh goes up to 1950Mhz at stock, while the max specification is 1455 MHz. So even though it's a 29% difference in clocks and that does give you a bit better performance, it does not represent 29% higher FPS while gaming, and the main reason is that we're limited by power.
 
Last edited:

Orpheus

Bronze Level Poster
Thanks a lot man, I followed the curve you sent me and it seems to be working well, no limit triggering whatsoever, 75°C max, these are really good results. I might tweak them a little bit later even though I wouldn't know where to start. Undervolting GPUs seems a bit more technical than CPUs somehow haha.

There's one thing that drives me crazy though, it's been happening for a few days now and I can't find a fix for it - and so my testing is cut short. After a relatively short period of playing time, the game would completely freeze for no reason and I'd have to manually shut it down with the task manager. The message that shows up is the following (translated from french) : "the application has been blocked from accessing graphics hardware" (or : "the application couldn't access graphics hardware").

Keep in mind that this was already happening before changing the curve, and I can't find any reason for it.
My GPU drivers are up to date, everything is fine but I don't know what's going on here. It's annoying as hell.


EDIT : Nevermind, I think the problem was my CPU undervolt after all, I bumped it back a little more and I didn't crash at all so far, let's hope it stays that way ! Thanks for everything, I'll def try to play around with the curve and see what I can achieve. But so far very pleased with the results, 75°C maximum for the GPU at 1775MHz (vs 86°C before). The only annoying thing is the fan noise but I can't do much about that, seems like they go off at 65°C.
 

Attachments

  • gfdhdth.png
    gfdhdth.png
    7.4 KB · Views: 187
Last edited:

FerrariVie

Super Star
Thanks a lot man, I followed the curve you sent me and it seems to be working well, no limit triggering whatsoever, 75°C max, these are really good results. I might tweak them a little bit later even though I wouldn't know where to start. Undervolting GPUs seems a bit more technical than CPUs somehow haha.

There's one thing that drives me crazy though, it's been happening for a few days now and I can't find a fix for it - and so my testing is cut short. After a relatively short period of playing time, the game would completely freeze for no reason and I'd have to manually shut it down with the task manager. The message that shows up is the following (translated from french) : "the application has been blocked from accessing graphics hardware" (or : "the application couldn't access graphics hardware").

Keep in mind that this was already happening before changing the curve, and I can't find any reason for it.
My GPU drivers are up to date, everything is fine but I don't know what's going on here. It's annoying as hell.


EDIT : Nevermind, I think the problem was my CPU undervolt after all, I bumped it back a little more and I didn't crash at all so far, let's hope it stays that way ! Thanks for everything, I'll def try to play around with the curve and see what I can achieve. But so far very pleased with the results, 75°C maximum for the GPU at 1775MHz (vs 86°C before). The only annoying thing is the fan noise but I can't do much about that, seems like they go off at 65°C.
Great that you figured that out and also that you're happy with the GPU undervolt/underclock results!

Next steps would be to save the current curve on one of the 5 profiles in Afterburner and start creating more. For example, if at some point you want the fans to be really quiet and you're not playing a demanding game, then just restore the stock curve and make a new one that flattens earlier, saving it as another different profile. Check the green example below (you could go even further than that, just play around with different values until you're happy with it):
1602230995084.png

Or in case you're not having enough FPS on very demanding games, you can make it flatten later. It will be close to stock speeds, while also producing a bit less heat (again do not consider the red line, I reused the screenshot from the previous post):
1602231101983.png

Then it's just a matter of applying your different profiles depending on your needs. That's what I do :) .

Regarding the noise, have you tried using a custom fan curve on Clevo's control centre? It does not have many options, but it might help you to not have the fans ramping up and down that much. I don't use Control Center anymore, but it should look like this:

1602231988667.png

I'm not saying that your curve should look like this, at it is just an example that I found on the internet. But you should again play around with the fan curve values until you're happy with it.
 

Orpheus

Bronze Level Poster
So technically, flattening the curve earlier means underclocking/undervolting, and flattening it later means overclocking/overvolting ?
Sorry if this is a dumb question, I'm used to tweaking CPUs but GPUs are a whole new thing to me.

Like, if my GPU goes up to 1V at full load, flattening the curve beyond this point would be risky right ?

Regarding the noise, have you tried using a custom fan curve on Clevo's control centre?
I did use it for the first few months after getting my laptop but found it to be very ineffective. The custom fan curve doesn't really work anyway, I used to set the fans to be silent until 70°C but it didn't do anything, they were still 100% spining at 60-65°C so I uninstalled it.
Seems like fan curves are BIOS locked on these Clevo machines.

Again, thanks for all the help man !
 
Top