Undervolting GPU: adjusting a voltage/frequency curve under 800 mV

[a_ill]

Hello!
I am using MSI Afterburner to undervolt my GPU (NVIDIA MX150). The problem is that it allows only to adjust values that are more than 800 mV. Is there any way to broaden the range or is there any other software that can be used to adjust values under 800 mV? Trying to make it as thermally efficient as possible.

 

[deathman20]

Hello!
I am using MSI Afterburner to undervolt my GPU (NVIDIA MX150). The problem is that it allows only to adjust values that are more than 800 mV. Is there any way to broaden the range or is there any other software that can be used to adjust values under 800 mV? Trying to make it as thermally efficient as possible.

Not sure its possible on this one but if you use the frequency chart, you can drag your "clocks" lower using the Alt-Shift method to drop it if using MSI Afterburner. Also you'll probably need to lower the TDP should push you into a sub 800mV range. With my higher powered cards it requires a 60-70% TDP to constantly stay that low in voltage.

Is there a specific reason with a card like that to drop it so low? Its already a low powered GPU to begin with, are you gaming on it at all? If not I'd suggest to even turn the thing off and use strictly onboard applications if that is the case, or utilize the optimus configuration and make sure it uses onboard basically for everything except when you require a GPU.

Edit: Now while my card is vastly different I have a Quadro M2000M in my work laptop. Barely has a boost frequency at least with mine since I run 3 monitors typically (2 desktop + 1 laptop) at my work station. Even with MSI AB I have zero voltage/TDP/fan controls. I can only increase/decrease my clock and memory speed a little bit. Course mine is an older gpu built on a 28nm revision.

 

[a_ill]

Not sure its possible on this one but if you use the frequency chart, you can drag your "clocks" lower using the Alt-Shift method to drop it if using MSI Afterburner. Also you'll probably need to lower the TDP should push you into a sub 800mV range. With my higher powered cards it requires a 60-70% TDP to constantly stay that low in voltage.

Is there a specific reason with a card like that to drop it so low? Its already a low powered GPU to begin with, are you gaming on it at all? If not I'd suggest to even turn the thing off and use strictly onboard applications if that is the case, or utilize the optimus configuration and make sure it uses onboard basically for everything except when you require a GPU.

Edit: Now while my card is vastly different I have a Quadro M2000M in my work laptop. Barely has a boost frequency at least with mine since I run 3 monitors typically (2 desktop + 1 laptop) at my work station. Even with MSI AB I have zero voltage/TDP/fan controls. I can only increase/decrease my clock and memory speed a little bit. Course mine is an older gpu built on a 28nm revision.

Unfortunately, dragging clocks even to the minimum doesn't make GPU use less than 800 mV. Power limit slider is greyed out and I haven't found any way yet to unlock it.

The reason is that I would like to run BOINC on it, don't like it just sitting there doing nothing. However, since I also run BOINC on my CPU, running it also on my GPU even at 800mV is going to make my laptop heat up more than 60 degree Celsius and I just don't feel comfortable with it. And no, I don't game on it.

I already use Optimus and iGPU handles everything except for GPU demanding programs, so Nvidia GPU is off all of the time.
This is how my voltage/frequency curve looks like (the top one is after adjusting).

I would like to find the minimal voltage that my GPU can run on (and therfore heat the least), but unfortunately I can't find any way to adjust lower power states.

 

[EarthDog]

You are worrying for absolutely no reason. 60c is more than acceptable man.... what temps does the system reach when using that gpu??? So long as it isnt throttling, its fine.

 

[a_ill]

I plan on sticking to my laptop for at least 5 years due to it being quite a capable machine, so I would rather not risk any component degradation due to high constant temperatures (running BOINC 24/7). I know that laptop components are more resilient, but still, do not want it to die on me. 66C CPU at 2.0 Ghz and 63C GPU at 1.5 Ghz while running 7 thread TN-Grid and Asteroid@Home, which is 6 degrees too much for my liking.

 

[EarthDog]

So, wait... your max temps with the GPU running are 66C and 63C? Dude, you are FIIIIIIIIIIIIIIIIIIIIIIIIIINE and will be for 5 years.

To be quite honest, if you were really worried about the laptop, you wouldn't be running its GPU and CPU at full tilt for BOINC. Those temps are nothing, not even close to nothing.

 

[a_ill]

Running them full tilt would be 3.4 and 1.7 Ghz accordingly and would send me into the world of >90C. So I am relatively cautious here

 

[EarthDog]

Relatively? I would say EXTREMELY cautious, like tinfoil hat paranoid.

And by full tilt I just meant the CPU and GPU running at 100%.

Well, you do what you do... We are saying the laptop will be just fine with much higher temperatures. But again, if you are truly worried about it lasting, reconsider running it at 'full tilt' for the next 5 years.

 

[Dolk]

You can't run a GPU under .8mV as this would lead to instability with in the transistors. Saturation voltage needs to be at least .8mV or greater.

 

[a_ill]

You can't run a GPU under .8mV as this would lead to instability with in the transistors. Saturation voltage needs to be at least .8mV or greater.

It runs at 0.675 V at 0.6 Ghz, so 0.8 V is not the floor. And I am sure that interval from 0.675 to 0.8 could be greatly optimised if I only had access to it.

 

[a_ill]

Undervolting CPU: adjusting a voltage/frequency curve

Hello!
As I know, it is possible to adjust a voltage/frequency curve for a GPU. However, I have not been able to find any topics about a voltage frequency curve adjustment for a CPU. Is it possible? If not, what is the reason?
I have an i5 8250U.

 

[Johan45]

Iread your other thread about the GPU and to each thier own bt personally I wouldn't be using a laptop for any kind of distributed computing. They aren't designed to run full out 24/7 and this will likely lead to premature death of the components.

 

[deathman20]

Speaking of temps for a laptop.... THAT is fantastic! No really thats great. Running anything under 75 is pretty darn good IMO. Now if you've already done it, raise the laptop up a little off the desk if running it that much and make sure its getting some nice airflow into the CPU/GPU area but really that is perfectly fine for it. Least on my laptop I OC the CPU and now might try OCing the GPU just for the heck of it.

Now as for the chart... ok that stinks about the TDP limit. I know when I've been playing around with my GPU (1060) to help drop the temp, I was able to tweak the OC on the Core to basically force it into a lower voltage range by dropping the TDP. While the chart is off to the side I've been able to get in the 750-780mV range constantly with my clocks for this card.

 

[a_ill]

Thank you for your reply. Even without running BOINC on it, I still would like to tinker with it and see how much more efficient I could make it. Optimising stuff is my thing. Also the topic itself is quite interesting to me.

 

[Johan45]

The CPU should do that on it's own, downclock and down volt.
Going to merge this with the other thread.

 

[a_ill]

Speaking of temps for a laptop.... THAT is fantastic! No really thats great. Running anything under 75 is pretty darn good IMO. Now if you've already done it, raise the laptop up a little off the desk if running it that much and make sure its getting some nice airflow into the CPU/GPU area but really that is perfectly fine for it. Least on my laptop I OC the CPU and now might try OCing the GPU just for the heck of it.

Now as for the chart... ok that stinks about the TDP limit. I know when I've been playing around with my GPU (1060) to help drop the temp, I was able to tweak the OC on the Core to basically force it into a lower voltage range by dropping the TDP. While the chart is off to the side I've been able to get in the 750-780mV range constantly with my clocks for this card.

How did you drop the TDP?

 

[a_ill]

The CPU should do that on it's own, downclock and down volt.
Going to merge this with the other thread.

Yes, it does that. But it follows the voltage/frequency pairs that were set by the manufacturer, which are far from ideal, as the graph above showed. As for CPU, I have been running it undevolted by 100 mV for several weeks and haven't encountered any instabilities yet, which shows that there should be as much room for optimisation as was for my GPU.

 

[deathman20]

How did you drop the TDP?

The TDP/Power Slider in MSI. Though if its locked most likely it doesn't allow for it. I know my graphics card on my laptop (Quattro M2000M) doesn't allow for any adjustment like that, just frequency and memory clocks so if yours doesn't have it, don't be shocked. Not sure what options there are on a laptop with GPU OCing / Tweaking. Totally new territory for me, wish I had a gamer card to try it on like the mobile versions of 1060, 1070 or 1080 Probably one of the few people that want to underclock stuff on a laptop.

Side note on the CPU at least since talking about it.
I just started playing with OCing/Tweaking my CPU on my work laptop so its new for me on this side at least, actually found out it was one of the first Intel CPU's that can be tweaked, the i7-6920HQ. Been able to up my CPU speed from the stock base of 2.9Ghz with a 3.8, 3.6, 3.5, 3.4 boost to the cores. To 4.0Ghz single and 3.7 for all following cores with a -0.1V adjustment. Higher clocks and lower temps, win win. Actually gives me about 5-6% more battery life with this tweak which I was very impressed with, and dropped my temps by 10C on the core when heavily used.

 

[a_ill]

Well, MX150 is actually just a mobile version of 1030. Haven't seen any people having problems with the power slider on 1030, so it's weird that it is locked on a mobile version.

Your undervolt + OC is one of the examples of an inefficient energy management by a manufacturer.

 

[Nebulous]

*sniff sniff*, what's that burning smell?