[Quick Guide] NVIDIA GPU Undervolting, Overclocking and Overvolting

I am seeing more people asking for help with undervolting their GPUs to improve some of the ludicrous thermals that are commonly seen in many of today's gaming notebooks. I made a quick video to help with this process for those needing assistance.

I also see a few folks now and then asking how to overclock their GPU. The same process used for undervolting can also be used for overclocking and overvolting. You simply go the other direction.

Rather than burying this information in a random thread where nobody can find it, I decided to make a new thread so it is easier to find using our forum search feature or Google search. Feel free to ask questions and share your successes in this thread. You will find links in the video description in my YouTube channel.



Edit: If you ever changed GPUs or vBIOS with Afterburner installed, you can end up with more than one .CFG file. If you have a multi-GPU system, you will have one for each GPU and all need to be modded. Sorry, I forgot to mention that in the video. If you only have one GPU and you are not sure which CFG file to edit based on the date of creation, just edit all of them and it won't matter on the file(s) not needed. Or, you can delete them all, let Afterburner create a new file(s) automatically, then edit the new file(s). But, on a single GPU system, Afterburner will create only one new file.

 

It's a nice little tutorial for people wondering how it should be done. There are tons of videos explaining this as well so it isn't really needed, however the thought is the important thing here.

 

Most people on laptops are likely gonna want the GPU to downclock at idle, so L or K-Boost probably isn't the best idea. I think a better way to undervolt is to move the whole curve up by adding an offset or dragging while holding down Shift, then flattening out. Like this:

Spoiler

 

Thank you for the compliment, bro.

This is a perfect example of what I meant by people sharing their successes. Thanks for posting it.

Yes, there are different ways to play with it and as people begin to tinker they can find what works best. Knowing where to start and getting past being intimidated by it is the hard part for some. To restore the ability to down clock at idle, simply restoring the default profile after you are done benching or gaming will take care of it.

 

@KY_BULLET - thanks for the reminder. Saw your post on YouTube.

Fixed...

 

No problem. My method is more of a set-and-forget method, so I'd imagine it will be more convenient for most people than changing profiles manually.

Afterburner has a setting in the profile tab as well to assign 2D and 3D profiles, so you could for example assign the stock dynamic curve to 2D and a locked undervolted/overclocked curve to 3D. But IME the automatic switching is not completely reliable, so I don't personally use it.

Lastly, @Robbo99999 has suggested in the past to undervolt by adding an overclock offset and then reducing the power limit. This works similarly to flattening out the curve, with the additional benefit of still being able hit higher clocks/voltage points under less demanding loads, but I'm not a fan of it because 1) It requires the ability to adjust power limit which mobile Pascal GPUs can't do without hardware flashing a modded vBIOS and 2) All Pascal mobile GPUs are already power starved by default and 3) Clock fluctuations from constantly bouncing off the power limit cause microstutter.

 

Updated the opening post with:

If you ever changed GPUs or vBIOS with Afterburner installed, you can end up with more than one .CFG file. If you have a multi-GPU system, you will have one for each GPU and all need to be modded. Sorry, I forgot to mention that in the video. If you only have one GPU and you are not sure which CFG file to edit based on the date of creation, just edit all of them and it won't matter on the file(s) not needed. Or, you can delete them all, let Afterburner create a new file(s) automatically, then edit the new file(s). But, on a single GPU system, Afterburner will create only one new file.

 

Mr fox why MSI afterburner compare to nvidia inspector? nvidia inspector looks noob friendly to me. Thats what im using.. ^_^

 

Nvidia Inspector is not as user friendly. You have to use command line for adjusting voltage/frequency curve, which Afterburner lets you do in GUI.

 

That's right, that has been my suggestion of mine, I don't remember seeing microstutter when it was bouncing off the power limit, maybe that's because I have G-sync, I might test it again to see if I can see microstutter. (I don't ever run my GPU like that though in normal use, because I have no power or temperature issues, I run it with max power levels, it was an acedemic exercise of curiosity.)

 

I have G-Sync as well, and I notice the small hitch when the clock speed changes by just one or two bins even in a non-interactive app like Unigine Heaven. Idk, maybe I'm just sensitive to it.

Do you typically use an FPS limiter like RTSS or in-engine? That could be why you don't see microstutter, as RTSS especially is known for helping smooth out frame times a lot.

 

@Mr. Fox are you saying it’s possible to overvolt pascal notebook cards to some degree or am I misunderstanding?

Thanks for the video.


Sent from my iPhone using Tapatalk

 

You're welcome.

Technically no, but in a practical application sense, yes. You cannot exceed the low voltage limit set in the cancer vBIOS, but you can force it to run full blast at the maximum for overclocking. By eliminating or reducing NVIDIA GPUs idiotic dynamic (spastic) performance and regaining a greater measure of control over the behavior of the GPU you can achieve better results, whether undervolting for better temps or maxing out the overclock at the voltage limit. The real reason we need videos for stuff like this is because the firmware architects are retarded control freaks.

 

Hi, I just retested to see if I could see any microstuttering - it was smooth as butter in Battlefield 1 campaign for 5 mins of play at a 50% power target and max overclock. The card was constantly bouncing off the power limit so the clocks were fluctuating the whole time (ranging from about 1797Mhz - 1898Mhz, with my normal clocks being 2050Mhz when I don't set a power limit), but fps was still over 100 fps all the time, ha and the temps wouldn't go over 60 degC and the fans kept coming on & switching off due to the low load! I attach here just a screenshot of my GPUz readout for the short play session:


Maybe you're talking from laptop experience of bouncing off the power limit/G-sync/microstutter - maybe the desktop cards & desktop G-sync can somehow cope with bouncing off the power limit better. I think I'm quite sensitive to microstutter, as I do most of my gaming at 144 fps in fast online multiplayer shooter games, so smoothness & predictibility is important to me. Yeah, I didn't see any microstutter when it was bouncing off the power limit.

I didn't answer one of your questions. I use an in-game fps limiter for Battelfield 1, set at 142 fps - this results in lowest possible latency. My test above it was only sometimes at 142 fps, but there were lots of time spans where the card was throttling due to bouncing off the Power Limit - those are the Green Sections seen in GPUz.

EDIT: this is only relevant for desktop cards or on the other hand mobile cards that have a modded vBIOS that allows you to change the power limit, otherwise I think Mr. Fox's video in the original post is the way to go.

 

I am somewhat the opposite. Unless it is severe, I normally don't notice or don't care that much if it is mild. But, I am more sensitive to stuttering than I am screen tearing. Unless it is really severe, I have to pay special attention and look for mild screen tearing or I never really notice it.

 

Maybe someone can find this helpful to improve their temps:


Note: Apply the undervolt values for both the gpu's then disable sync function in MSI.

Only tested on Rise of the Tomb Raider and saw a 5 deg c decrease in temps during game play.

Ultra Settings 3k resolution
De-lidded CPU running at 4500 (High performance -Power Management)
Nvidia Power Management - Maximum Performance
Modded MC U3 Plus with 3000 rpm fans running at 70%
Liquid Metal for both CPU & GPU
K5 PRO for GPU VRM & Chokes

 

Nice! Thanks for sharing this.

 

Yeah my only experience with Pascal and G-Sync so far is on a laptop. I've seen other laptop Pascal/G-Sync users also claim stuttering when clocks fluctuate. If you're not seeing it, defnitely could be down to differences between mobile and desktop G-Sync (maybe the G-Sync module helps). Did you try disabling G-Sync and checking for stutter?

 

That's quite a sharp rise in your curve at the circled part. You lose almost 100MHz between 943mV and 933mV and about 50MHz between the other points. Should you ever hit a power or thermal limit, you'll lose a lot of performance because your clocks will drop like a brick when it switches to the lower voltage points. You should keep a gentle slope in your curve like this:

 

I might try that later.

 

You're right but the undervolt for 1784 & 1733 range doesn't really matter for me since my system maintains the 950 voltage and the clock speed never went below 1800 mhz during 1hr of game play in ROTR. This where I think the decrease in temps happened since in stock both gpu's are running at 1.05 or 1.06 V and boosting to 1800-1873 range while in undervolt value I'm getting the same Mhz even boosting to 1898 for few seconds lol. I'll be doing more testing also with your setting

 

Hi, got a question about MSI Afterburner. Sorry if its a noob question but its my first time using the program. Why is it that MSI Afterburner GPU Memory clock is showing doubled compare to GPUZ. Like my GTX 970m has 1252.5 MHZ GPU Memory clock showing in GPU Z and in MSI Afterburner it is showing as 2505 MHZ. Also is there a way to make MSI afterburner show the same GPU memory clock as GPU Z?

 

That's normal and nothing to worry about. The correct actual frequency of the VRAM is shown in GPUz. Because GDDR5 is "quad-pumped" some websites & programs will multiply the actual frequency by x4 to provide an 'effective frequency', and other programs choose to multiply the actual frequency by x2 (like MSI Afterburner) - it's actually a pretty silly thing for them to do, they should just report the actual frequency in my opinion, so kudos to GPUz for telling it like it is!

 

Thank you very much for the info..

 

Select "memory clock" and enter "x/2" inside this box:

 

Cool.. I'll try this later as I am currently out.. Just want to add 1 more question.. You know in Nvidia Inspector you have a slider to. increase and decrease the clock speed and also it has a button to increase or decrease by 1, 10, 20 and etc. For MSI Afterburner does it have one that is the same or something to click 1 by 1 to slowing increase or decrease value? This morning before leaving the house I tried messing about with the program and tried some stress test with new OC value.. When OCCT and Heaven benchmark was running, I tried changing the value by typing what value I wanted and suddenly it changed itseld to +700 plus Core clock speed.. My laptop froze and I rushed turning it off manually.. LOL.. so yea, does MSI Afterburner have a setting to increase value by clicking 1 by 1 and slowly but surely.. Hope this is understandable as my english is not that great.. Thanks in advance..

 

(Pascal core frequency operates in step frequencies of 13Mhz, so increasing by values of 1Mhz will do nothing until you get to the next 'step' up. Memory frequencies can be increased by 1Mhz if I remember rightly.)

 

And, the effective (actual) memory speed for GDDR5X is octa-pumped, so multiply x8. I really do not understand why everyone in control of this stuff doesn't get with the program. We don't say our DDR4-4000 is 2000MHz. Maybe they just like to confuse people that don't know the readings are not displayed correctly. Or, maybe it is too complex and requires too much time consuming effort to fix software that is distributed to everyone for free.

 

You can move the slider, or enter values manually, and click apply. Maxwell isn't locked to 12.5MHz increments on the core like Pascal is.

 

The core clocks on 980M and the 180W and 200W GTX 980 cards that I have had only moved in 13Mhz increments. I'm pressing my memory now, yet I vaguely remember that Kepler did as well... but, that was too long ago to remember clearly. My brain tends to purge things that no longer matter.

 

Interesting. Both my 980M (Prema vBIOS) and Kepler 650M (svl7 vBIOS) did not move in 13MHz increments.

 

+3

I mean my 780M with (svl7 vBIOS) moved in 13MHz increments. Not 100% sure but almost.

 

I think it matters whether GPU Boost is disabled in the vBIOS.

 

It was that way for me with both @svl7 and @Prema vBIOS from Kepler forward as best I can remember. Fermi was too long ago for me to even pretend to remember clearly.

I have been using modded vBIOS on NVIDIA and AMD graphics cards since 5870M and 580M, so I would not remember differences between stock and modded vBIOS for any of them, other than stock firmware always sucks.

 

This worked.. Thank you very much..

Yea, its not the same with nvidia inspector that it has a arrow on the side of the slider so you can click and increase slowly. Never mind I'll just use nvidia inspector to test and MIS Afterburner for final clocks and save profile..

Thank you for opening this thread.. Since its coming from you who is very credible, you've basically encourage me to try the program.. Thank you again..
Lastly Mr. @Papusan who has been teaching me about overclock in the back ground.. Thank you.. ^_^
As of the moment I have overclock my 970m to +198 on GPU core clock and +291 Mem clock.. This is on Stock voltage.. Next would be increasing voltage and OC..
Anyone got any rough idea whats the safe daily use voltage of 970m. I was thinking 1.025V? or just 1V?

Edited:
How come in MSI Afterburner I cannot change the voltage even though I am already using prema's vbios? In Nvidia Inspector there is no problem..
Edited again:
The answer is in Mr. Fox Vid.. Sorry.. haha..

 

As long as you keep the temps in check, have proper thermal pads fits on components (only graphics core have temp sensor) I mean you shouldn't have issues with this voltage levels(+1.025V). But I'm not sure how well the VRM's is for your BGA graphics. But it's well known that several 980m cards went to hell even with stock voltage (not sure about 970m). Not due fried cores, but the VRM's couldn't keep up. Bro @Mr. Fox had 2 or was it 3 of his Gtx980m who died. To have it in alive he added one or two missing VRM's.

 

Thank for the info. Yea the prob is the VRM. I am still using a stock pads as I am too lazy to do a repaste again as my temps are very good with gelid extreme. Max Cpu temp when running Heaven Benchmark and OCCT for 1 hour with OC settings is 71 degree celcius. For GPU max temp was 67 degree celcius. My plan for my max voltage with this GPU is only 1.0120V and CPU and GPU temp must not reach 75 degree celcius or more. Hope this plan is safe enough. I guess the only way to find out is try it. hahahahahha.. Anyways I am very happy learning something new.. MSI Afterburner.. hahahahha

 

@Dr. AMK got my golden 980M in the P750ZM he purchased from me. It was heavily hardware modded and overclocked like a banshee. It was beefed up by @Khenglish to keep it going strong.

 

Yes, it's the most powerful 15" laptop I ever have. Now the new owner is a 3D and CAD designer and he is very satisfied with this laptop performance with his rendering activities.

 

Thumbs up to @Zoltan@HIDevolution for pointing me to @Mr. Fox video. Thanks to this guide I've been able to undervolt the GTX 1080 in P775TM1 to 912mv @1847Mhz. My idle temp both before and after undervolting remains at almost a constant 57C. I tried out Unigine Superposition for about 15 minutes and it reached around 83C. Is that "good enough"?

 

Sweet! Glad to hear it helped. That is excellent.

 

@Mr. Fox Hi, me again. I have noticed that when I open MSI Afterburner with OSD and play PUBG I get micro stuttering and massive FPS drop. This happens when I move the mouse cursor and not even playing the game, just in the lobby. Also in GPU Z gpu clock speed and mem speed throttles down.. This is only happening on MSI Afterburner. I had to go back to Nvidia Inspector which has no issues. Got any idea whats causing this? Also what is the difference between Nvidia Inspector and Nvidia Profile Inspector? It seems Nvidia Profile Inspector has been updated and Nvidia Inspector has not had any update in a long time. Got any idea why Nvidia Inspector has not been updated? Thanks for your help.

 

Howdy, bro.

NVIDIA Profile Inspector is for advanced editing of GPU driver profiles and exporting them. Great tool, but it has nothing to do with your situation. It is not for overclocking. It's mainly for games and it saves time having to tweak things in NVIDIA Control Panel every time you change drivers. Just import the profile and you're done with the tweaking. I can only guess, but my assumption is that NVIDIA Inspector (the overclocking tool) has not been updated simply because no update is needed. It still works as is, so why bother with an update? Being old is of no consequence if it still works. An update might screw it up, LOL.

Go ahead and use MSI Afterburner to set your clocks and voltage as shown in the opening post. Then shut down MSI Afterburner. It does not need to be running once the settings are applied. They should stick until your reboot. Try that and see if PUBG behaves correctly. If it does not behave correctly after closing Afterburner, then the problem is most likely the settings you are applying using Afterburner rather than Afterburner itself.

 

If you check this setting, Afterburner will automatically apply your clocks after login even if it's not running:

Spoiler

You can't use Nvidia Inspector to adjust the voltage/frequency curve like you can with Afterburner. (Well technically you can, using command line, but it's a b*tch to do without a GUI).

 

This problem is caused by horrible Ngreedia drivers.
Downgrade your drivers.
390.77 and 391.X seem to be stable, as well as 388.xx.

PUBG also have bugs with TDP unlocked GPU's, causing power limit throttling for no reason even when you are nowhere remotely close to the power limit (probably has something to do with the 144 fps cap doing something funky). But I have absolutely NO stuttering with mouse and MSI afterburner/RTSS / HWinfo64 OSD in PUBG main menu.

Now, Xbox 1 with Bluetooth controller is a different story. Don't get me started.

 

Yes, that works fine and I have used it many times. But, (yes there is always a "but" to consider,) you need to be done with testing and be 100% certain the settings are stable and you did not have something else applied at shutdown. If you have that selected and you are still testing unknown settings that may not be stable, it can do more harm than good. If it crashes with wrong settings, having them automatically applied at Startup requires booting into Safe Mode to undo the mistake. If you save the known good Afterburner profile and apply it from the system tray icon whenever you actually need to use it, that is a safer option. Nothing gets applied by accident if you boot using GPU defaults and manually set the profile from the system tray.

 

Spoiler

 

I have actually had that not work before. Not sure why, but I once needed it to work and when it didn't I had to boot into Safe Mode and temporarily rename the Afterburner exe and cfg file to boot into Windows without the screen going black from borked overclock settings being applied at logon.

Without BCD tweaks, getting W10 into Safe Mode can be a real pain in the butt sometimes.

 

Thanks for the reply guys. I just tested what Mr. Fox said and my OC profile remains on once activated even though I have close MSI Afterburner. I have found the culprit. It was RivaTuner. When I uninstalled it, it fixed the problem even though MSI Afterburner remains open. This didn't happened before. I am assuming its due to PUBG on going patch and updates which caused instability with RivaTuner. Sucks I really enjoyed the OSD as I am new to this. hahaha.. Too bad it only lasted for a month or so.. Mr. Fox is there a alternative software for RivaTuner's OSD?

Thanks for this but I am currently using 390.77 which is really stable for me..

 

Not a good one. Try changing the RTSS settings and that might help. Toggle some of the options in RTSS.

If you are outputting too much info to the OSD overlay, that can also cause issues sometimes.