Undervolting a 1660Ti

BoomAM · 5 Aug 2019 at 23:37

Hi,
I just got a 1660Ti for a secondary PC, sat in a Phantec's Evolv Shift.
Doesn't run too hot, but i want to make it a little cooler to reduce fan noise, hence, undervolting.

I'm trying to follow some guides that are posted on LTT or Reddit in regards to the process, but it makes no sense whatsoever.
Using MSI Afterburner, i can adjust the pretty little graph that shows, BUT most of the guides list dragging the line DOWN to adjust the voltage. The up/down axis is frequency however?

Am i missing something obvious here?

Thanks in advance!

HeX · 6 Aug 2019 at 00:16

Vertical is frequency, horizontal is voltage, just make the curve stop at the voltage/frequency you want then flatline it. Then test if it's stable, adjust as needed.

BoomAM · 6 Aug 2019 at 00:24

HeX said:
Vertical is frequency, horizontal is voltage, just make the curve stop at the voltage/frequency you want then flatline it. Then test if it's stable, adjust as needed.

Yes, thats what i said.
All the videos/guides online show people adjust the frequency to lower the voltage!?

Interestingly, i'm playing around with increasing core frequency on the main MSI menu, whilst limiting the max power.
Running a bench in Heaven benchmark and recording max temps/score.
Stock = 2245 points, max temp of 75.
+100Mhz core, -10% power = 2285, max temp of 72
+150Mhz core, -20% power, = 2319, max temp of 71

Makes no sense at all other than 'because nVidia boost', but ok, i'll take it.

Poneros · 6 Aug 2019 at 00:29

It's because frequency & voltage are interlinked.

BoomAM · 6 Aug 2019 at 01:05

Poneros said:
It's because frequency & voltage are interlinked.

In what way? As the results im seeing seem to be a bit like a 'yo-yo'.

Just replaced the stock Zotac paste with some MX4. Lets see if that also helps.

Poneros · 6 Aug 2019 at 01:28

BoomAM said:
In what way? As the results im seeing seem to be a bit like a 'yo-yo'.

Just replaced the stock Zotac paste with some MX4. Lets see if that also helps.

In the sense of frequency being linked with voltage. So if you wanted a higher voltage, let's say, for a particular frequency, then you couldn't do that. There's a particular voltage & frequency pairing and you can't modify them, you can only select them. So if you want to choose a particular voltage, you can (for the most part), do that. But you can't then also choose which frequency to go with that voltage. And vice-versa. So you can choose voltage, or you can choose frequency, but not both at the same time. Whichever you choose, the second is given to you.

BoomAM · 6 Aug 2019 at 01:34

Poneros said:
In the sense of frequency being linked with voltage. So if you wanted a higher voltage, let's say, for a particular frequency, then you couldn't do that. There's a particular voltage & frequency pairing and you can't modify them, you can only select them. So if you want to choose a particular voltage, you can (for the most part), do that. But you can't then also choose which frequency to go with that voltage. And vice-versa. So you can choose voltage, or you can choose frequency, but not both at the same time. Whichever you choose, the second is given to you.

ok, so how does that help me undervolt? What is the correlation and how would this apply in practice within MSI afterburner?

HeX · 6 Aug 2019 at 10:04

It's just a case of trial and error to find the sweet spot for frequency at whatever voltage you are wanting to undervolt to.

Say you want to top out at 975mV, then you'd just drag every point after 975mV down to the frequency linked with that voltage. This is the known working frequency at that voltage level. You can then try upping the frequency at this voltage until you find the point of instability, then just back it off a bit and your card will now only boost to this new max level at the set voltage.

Bluntwrapped · 6 Aug 2019 at 12:17

For undervolting the beginning of the curve needs to begin lower down or further to the left of the screen. The upper most part of the curve, where it levels out is where the voltage tops out.

One thing to do though is benchmark before and after the voltage drop. The card may still be stable after, but the performance can drop if you’re right on the threshold of stability.

varkanoid · 6 Aug 2019 at 13:58

BoomAM said:
Just replaced the stock Zotac paste with some MX4.

I guess you didnt care about the warranty ?

SkeeterUK · 6 Aug 2019 at 14:13

Why would it void warranty if its the void warranty if removed sticker that your worried about then that was in court proven to be unenforceable and to ignore such stickers, a few years ago.

clown_abhi · 6 Aug 2019 at 15:15

SkeeterUK said:
Why would it void warranty if its the void warranty if removed sticker that your worried about then that was in court proven to be unenforceable and to ignore such stickers, a few years ago.

Yeah but the companies dont listen to the court and do you have the time to convince them otherwise?

clown_abhi · 6 Aug 2019 at 15:21

BoomAM said:
Yes, thats what i said.
All the videos/guides online show people adjust the frequency to lower the voltage!?

Interestingly, i'm playing around with increasing core frequency on the main MSI menu, whilst limiting the max power.
Running a bench in Heaven benchmark and recording max temps/score.
Stock = 2245 points, max temp of 75.
+100Mhz core, -10% power = 2285, max temp of 72
+150Mhz core, -20% power, = 2319, max temp of 71

Makes no sense at all other than 'because nVidia boost', but ok, i'll take it.

Wont reducing power reduce performance as well? I have set power at the max possible which goes to 108 on my card and it reaches max temp of 72 and sometimes 74 when ambient is hot.

BoomAM · 6 Aug 2019 at 23:26

HeX said:
...then you'd just drag every point after 975mV down to the frequency linked with that voltage...

That's kinda the reason I've posted to ask this question.
Dragging down is the frequency, not the voltage. You and everyone else are seemingly referring to the wrong axis. This makes no sense to 'drag everything after 975mV down'.
So either its badly explained as a whole and the 'community' around this aspect has just taken a bad description as the accepted explanation, or theres some strange axis swapping going on with my system here. ;-)

(Note, not having a go at you. :-)

)

varkanoid said:
I guess you didnt care about the warranty ?

Not really.
Plus, there were no 'void if broken' stickers to be seen ;-)

clown_abhi said:
Wont reducing power reduce performance as well?

You'd think, but from the sketchy assumptions im seeing online regarding the nVidia boost tech on these, sometimes it actually helps. Case in point, what i'm seeing.

I'm seeing some good results with MSI afterburner set to a max power rating of 80%, +150 on the core.
After letting heaven benchmark run for about 15-20mins in a 24*c room (75*f) the max temp seen is about 68*c.
Gonna run it for a bit longer, but if this method appears to work (higher perf, lower temp) then the faffing with the graph could ultimately be pointless.

SkeeterUK · 6 Aug 2019 at 23:46

Dont use heaven for testing overclocks etc i hear its no good for that as you need something more precise or something like firestrike.

BoomAM · 7 Aug 2019 at 00:12

SkeeterUK said:
Dont use heaven for testing overclocks etc i hear its no good for that as you need something more precise or something like firestrike.

Noted, but im not testing for out and out performance, just to check performance regressions...

Does anyone know how the 'set clocks on startup' function in MSI afterburner works?
Do i need to manually open the app on each windows boot? As i cant see a service running to set it on Windows boot...

HeX · 7 Aug 2019 at 07:29

BoomAM said:
That's kinda the reason I've posted to ask this question.
Dragging down is the frequency, not the voltage. You and everyone else are seemingly referring to the wrong axis. This makes no sense to 'drag everything after 975mV down'.
So either its badly explained as a whole and the 'community' around this aspect has just taken a bad description as the accepted explanation, or theres some strange axis swapping going on with my system here.
(Note, not having a go at you. )

Yes vertical is frequency, all your telling the card is don't boost any higher than X frequency at Y voltage, so every boost data point past the one you want is irrelevant.

If for example you set something (totally made up numbers here) like this boost table:

800mV - 1000MHz
850mV - 1100MHz
900mV - 1200MHz
950mV - 1300MHz
1000mV - 1300MHz
1050mV -1300MHz

This would give a volt/frequency curve that flatlines as described. The card would boost to 1300@950mV and ignore the higher voltage values as it's already achieved that boost clock at a lower setting.

You don't even have to flatline it, you can set them to anything as long as they are the same/lower than the frequency/voltage combo you want.

clown_abhi · 7 Aug 2019 at 09:01

When i was originally doing testing on the overclocks, i had to set power to 108 to get a stable overclock of +150 on the core and +500 on the memory.

If i bring the power down, the overclock wont work anymore. I guess its either higher clocks and more power or lower clocks and lower power.

I will try to set it to about 90% power and see what happens.

Does memory overclock give any benefits?

varkanoid · 7 Aug 2019 at 10:08

BoomAM said:
Not really.
Plus, there were no 'void if broken' stickers to be seen

Doesnt matter Zotac know when its been removed from experience. They know you've undone the screws.

BoomAM said:
Noted, but im not testing for out and out performance, just to check performance regressions...

Does anyone know how the 'set clocks on startup' function in MSI afterburner works?
Do i need to manually open the app on each windows boot? As i cant see a service running to set it on Windows boot...

Tick the startup option in settings so it starts with Windows.

Ideally to prove a point you needed to run a benchmark at stock then alter your intended settings then run same benchmark and show the results here then we can see if they have made any difference.

Use 3D Mark Timespy or Unigine Superposition these will show up good results.

clown_abhi · 7 Aug 2019 at 15:42

varkanoid said:
Doesnt matter Zotac know when its been removed from experience. They know you've undone the screws.

Tick the startup option in settings so it starts with Windows.

Ideally to prove a point you needed to run a benchmark at stock then alter your intended settings then run same benchmark and show the results here then we can see if they have made any difference.

Use 3D Mark Timespy or Unigine Superposition these will show up good results.

I have done the starup option thing but sometimes it applies the overclocks sometimes it doesnt and i manually tick the profile and apply.