Confused about GPU undervolting in MSI Afterburner (laptop 1060)

While CPU undervolting (via XTU or Throttlestop) in laptops seems like a popular subject, there's much less info to be found on GPU undervolting. I've been tinkering with my new Gigabyte Aero 15 with a 1060 and I've decided to give GPU undervolting a go.

The general idea is as follows: open Afterburner, hit Ctrl+F, adjust points on the curve so that lower voltage is hit at a given frequency. Seems simple enough.

Example (laptop 1080):
before
after

Now for my confusion.

This is my 1060's default curve.
This is my 1060's curve after a simple +200 core overclock (I haven't touched the curve manually at all).

Forgive (and please enlighten) me if I'm dense, but haven't I just undervolted the card by overclocking it? All points on the curve are shifted up, which should mean the card uses less voltage at any given frequency point.

If so, are overclocking and undervolting really 2 sides of the same coin when it comes to GPUs? Is there any point in further tinkering here, or should I just leave it at +200 and call it a day?

 

Nope your confusion is justified.

Technically, undervolting would be shifting the curve left. But the core offset only shifts it up or down which means simultaneously:

1. the card will run a lower voltage for a given clock speed
2. the maximum clock speed is increased

If you only want to undervolt, you need to choose a point and set the curve to be flat to the right of it. This will set a frequency/voltage cap.

It's a complete pain to do it though... The interface is sooooooo clunky. I wish curve point values were stored in plain text in the profile file but it's a 6000 byte long numerical string...

 

Here are some how to undervolt CPU / GPU videos I posted earlier:

http://forum.notebookreview.com/thr...scussion-thread.794965/page-428#post-10630086

http://forum.notebookreview.com/thr...scussion-thread.794965/page-428#post-10630138

Sorry, I haven't used the undervolt GPU stuff, only OC'd GPU's, which as you noted is similar to undervolting by using the excess voltage instead of burning it off - keeping the same voltage at higher performance demand. But, that increases thermal generation - and performance - rather than reducing thermal generation at the same performance.

When you get that aha moment and get it undervolted putting out less thermal energy, please do enlighten us on how it works

 

Thanks guys, it's clearer now.

I think I'll stick to just overclocking then, since the laptop isn't throttling and the fans don't spin up that much either.

 

As @bennyg said.
Pascal is a complete mess. Not only can you not directly undervolt, there are two other issues:
1) overclocking the core too high can just make the card crash.
2) having the core run at too low a voltage (not exactly the same as overclocking) can cause decreased performance without it even making sense.

You should check some of the videos from that guy who made "The Pascal Problem" youtube video, but I think the only thing you can do is to set the curve to be completely flat if you wanted to OVERCLOCK at a fixed frequency at all ranges. This worked on a modded DESKTOP card which had soldered components to allow more than 1.1v of voltage via a potentiometer, but all you would accomplish on a laptop card is just create more spectacular ways for it to crash. Otherwise do what bennyg said and find a point on the curve and set all points to the right to match the same frequency (flat out).

 

See, this is what I can't get my head around. If my overclocked card is maxing out at under 1800 MHz, it's not going to exceed 850-ish mV anyway, so what's the benefit of "flattening" the curve to the right?

 

The thing is, this works better for the 1070 or 1080 that will consistently pull more volts, while the 1060 is frequently pulling below the minimum in the voltage curve in AB of 800mV.

 

For me? My vortex MSI 1070s start crashing at over 1.013V

I'm also running them on undersized heatsinks so every bit of power saving helps. The 1070 being so power starved at stock, you can guarantee anytime its 100% utilised it'll be power throttling well down the curve, and if it's not 100% utilised - what is the point of clocking it higher with more volts, just making more heat? Its going to be completing each frame (slightly) faster but if that's not the bottleneck, why bother? It kind of makes me think of highway cruising in a higher gear than needed - extra rpms for nothing.

 

Vortex?
If your MSI MXM 1070 is crashing at 1.013v and above, then you have the old revision MSI 1070 that came out for the skylake GT72VR and GT73VR (6820HK). That is a hardware flaw on the card that no one figured out what part is to blame for this failure (otherwise someone could just solder a part on, if they had the expertise to do so).

I personally have had no problem with the card running at 1.062v and 2088 mhz (only got this speed if the temps are below 42C).

 

Originally from Vortex G65VR (6RF/7RF SKU's), the expensive "trash can" pc

Only them and the GT73VR/GT83VR1070's have SLI ports. The voltage crash bug is why they were discounted on ebay.

I use a 1975mhz/0.95V profile which could be tweaked further but that is still limited at 150W and I'm already limited enough by temps anyway.