Soldato
Might be a bug always worked when I used to use it. Only time you had to be wary is when you update the driver as it removes the overclock until you reboot.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Undervolting a 1660Ti
- Thread starter BoomAM
- Start date
Might be a bug always worked when I used to use it. Only time you had to be wary is when you update the driver as it removes the overclock until you reboot.
Associate
- Joined
- 22 Nov 2009
- Posts
- 106
Thank you. Didnt know that. Will keep it in mind
Im just gonna give up on this option.
Whilst i understand the concept, the process of doing it makes no common sense for how a program is laid out to achieve it.
I'll take the risk.
I know this.
I was asking how it set it. Usually things like this set with an app or service on boot. Neither of these are present on the system.
...above, Post #3.
Soldato
I know this.
I was asking how it set it. Usually things like this set with an app or service on boot. Neither of these are present on the system.
...above, Post #3.
Actually you are incorrect usually startup actions are set in the registry startup key or the windows startup folder.
You were advised to use a different benchmark other than Heaven hence post the results. The ones you have posted are not very conclusive as you could run Heaven 10 times with the exact same settings everytime and you can get a differentiation of +/-100 points. The conclusion from your initial test just proves the settings you have changed to make very little difference because the GPU is coping with the alterations. To prove your point you would need to change the settings more dramatically. It just shows the 1660ti you have is a good overclocker if you can reduce power by 20% and +150 core and it not crash but you need to record the boost clock too. However as been said you need to use at least one other benchmark. I've had overclocks that pass Heaven but then fail miserably in 3dMark.
Soldato
It's just a case of dragging the dots up and down.
In the example below we want the card to use a max voltage of 1000mV, as such we can drag down every point after 1000mV to the same level. The GPU will then ignore every point after this as it's already at this boost level.
However, this would give an undervolt AND an underclock as we're telling the card to stop boosting at a lower level.
If you want the card to undervolt AND boost to the same level, then what you need to do is raise the frequency clock on a lower voltage state then flatline after that point. In the example below we've set the full boost frequency at the 1000mV point, and then flatlined beyond that.
The card will now boost to it's stock speeds at a lower voltage and ignore the voltage points beyond as it's already at the max boost level.
Thats not accurate - in the case of physical hardware related settings, usually these are set via an app that accesses the hardware's address bus in-lieu.
This is because Windows lacks the correct components/APIs to be able to talk to custom hardware buses, a registry entry would not allow that as its just a variable for an application to store/check a setting when its running.
If it has been set in the registry, i would have guessed that its just a call function for the service/app that runs in the background that does know how to access the hardware directly, with said key telling the app what to do.
The only way this is achieved without an accompanying app, is if the nVidia driver is providing that link to the hardware and recognizes the registry setting as something it needs to set on its own to the physical hardware on boot.
....which is what i was asking, how its achieved at software level.
Whilst i see what you are getting at here, the performance regression testing was duplicatable each time. Using a more stressful test would not have improved that outcome.
By most testing methodologies, that's conclusive evidence.
Yes, that does appear to be one of the conclusive results. Which is surprising considering how new the 1660Ti silicon is to the market. I either hit the silicon lottery or the yields are efficient already.
IIRC, boost clocks were sustaining above 2Ghz, which is noticeably above 'stock' boost clocks.
I've not read all of your post yet, i will give it a good read when i get in tonight, but to pre-post as it were, thank you for providing, it looks useful.
I will read & test when i get in.
Still not read HeX's post, but to add to this, my other half (who owns this 1660Ti) has decided she wants to WC the thing. So we're going out to get a G12/an AIO that'll fit in her SFF. 
So this whole thread could be practically pointless, even if productive in theory.
So this whole thread could be practically pointless, even if productive in theory.
Soldato
Water cooling a 1660Ti, an already very cool running card that'll hit the limits of the silicon on air, is madness!
Tbh, its the case thats the issue.
Phantec Evolve Shift (non-X) - its airflow starved, so its a miracle i can get it running at 70*c with that small tweak, but the fans hit 70-75% and is loud.
We're literally gonna replace the H80i v2 for the CPU, with 2x H55's instead as they should both fit in. With the aim of sacrificing some CPU *C for an overall quieter system.
Turns out the local store didn't have the H55 in stock....so before we order online, i played around a little more...
Other observations, a locked fan speed of 40% shows the highest temp to be 77*c.
I've now set a custom fan curve that locks then fan at 40% upto 70*c, then ramps up to 50% at 75*c and 100% at 80*c.
That should tide things over for a while, whilst t'other half starts getting into gaming proper. If it ends up being an annoyance, we always have the H55 option (and i'll take the 'old' CPU AIO and whack it on my 1080Ti)
- 70% power limit.
- CPU fan locked at 50%
- CPU core boost of +175Mhz
- Max temp seen: 73*c
- Avg temp seen: 70*c
- Max core speed seen: 1945 Mhz
- Avg core speed seen: 1775 Mhz
Other observations, a locked fan speed of 40% shows the highest temp to be 77*c.
I've now set a custom fan curve that locks then fan at 40% upto 70*c, then ramps up to 50% at 75*c and 100% at 80*c.
That should tide things over for a while, whilst t'other half starts getting into gaming proper. If it ends up being an annoyance, we always have the H55 option (and i'll take the 'old' CPU AIO and whack it on my 1080Ti
)
Just revisiting this.
Is this just a case of your 'first' dot, adjusted as the max clock speed at a given voltage on the horizontal axis, becoming the new max voltage? Assuming that the following 'dots' in the graph are on the same level on the vertical axis?
So if i set for example, 1,000Mhz to be 0.5v, and everything after that is set to be at the same horizontal 'level' as that, then the max voltage would be 0.5v at any given speed above 1,000Mhz? With the speed maxing out at the best it can achieve at that voltage?
If so, then that is understandable, but a completely messed up way for MSI Afterburner to represent that.
So i tried dropping the voltages at 2Ghz down to 0.950 and 0.900, in both scenarios i got more heat output?
(done by finding the voltage, moving the dot up to the frequency, then flattening everything after that to the same level).
Tried combining adjusting the voltages charts with the max power limit and instead saw a 200-300Mhz drop in clock speed, and still, more heat.
Went back to my original 'basic' undervolt (40% locked fan speed, +175Mhz core, 70% power limit) and i got back to good temps again and higher/more consistent 1,950-2,050Mhz clocks again.
No idea what is going on there! Especially considering the voltages with the basic undervolt were a little higher.
(done by finding the voltage, moving the dot up to the frequency, then flattening everything after that to the same level).
Tried combining adjusting the voltages charts with the max power limit and instead saw a 200-300Mhz drop in clock speed, and still, more heat.
Went back to my original 'basic' undervolt (40% locked fan speed, +175Mhz core, 70% power limit) and i got back to good temps again and higher/more consistent 1,950-2,050Mhz clocks again.
No idea what is going on there! Especially considering the voltages with the basic undervolt were a little higher.
Is there a trick to flatten curves in Afterburner? Like a shortcut?
That's a shame, as its as PITA to do manually.
Anyway latest result seems good:
0.800v max voltage
1,800Mhz frequency base
Custom stepped fan curve.
Had 'other half play the GPU killer itself, the title that puts Furmark, Heaven, SuperPosition & Vally to shame when it comes to to heating the 1660Ti up....Snake Pass.
Max temp we saw was 65*c, and the fans mainly ran at 45%, which from 7-8ft away on the couch isnt really noticeable. This was after 30-40mins of playing.
I'm reasonably happy with that...
Anyway latest result seems good:
0.800v max voltage
1,800Mhz frequency base
Custom stepped fan curve.
Had 'other half play the GPU killer itself, the title that puts Furmark, Heaven, SuperPosition & Vally to shame when it comes to to heating the 1660Ti up....Snake Pass.
Max temp we saw was 65*c, and the fans mainly ran at 45%, which from 7-8ft away on the couch isnt really noticeable. This was after 30-40mins of playing.
I'm reasonably happy with that...
Soldato
Doesn't have to be perfectly flat, as long as the values are the same/lower than the undervolt you want you're fine.
its not the overall flatness that's the issue, but when every point past 0.800v is higher, i have to adjust each one.
With the results i'm getting out of tuning the 1660Ti, i'm half tempted to give the same undervolting a go with my 1080Ti.