• 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.

30xx Series Founders Edition

Associate
Joined
23 Aug 2005
Posts
1,273
What clock speed are you using at 0.9v and how does it bench compared to standard?
I've set it to 1900Mhz, GPU-Z reports 1890MHz. I wanted to try lower volts but can't seem to set it lower than 0.9v in MSI Afterburner, maybe if I reduced the clock speed. Standard clock is 2100MHz I think.

Superposition benchmark at 1440p Extreme/High/everything enabled: Score 4932 FPS: Min 29.52, Avg 36.89 w/3600. Others overclocking the card can get up to 5452, Avg 40fps.
 
Associate
Joined
10 May 2012
Posts
216
Location
Leeds
Has anyone had any issues with their monitor not going into power save mode when sleeping. Never had this problem with my old GPU and that was using DisplayPort as well.

I've got 2 monitors connected 1 DisplayPort and 1 HDMI, the HDMI monitor goes to sleep, the DisplayPort monitor goes black for a few seconds, you then hear the Windows device unplugged sound and then the backlight of the DisplayPort monitor comes on and the screen stays black, it then stays like that until you move the mouse.
Both monitors should be going off though.

Bought a VESA Certified DisplayPort cable just in case it was a crappy cable, but same issue.

[Edit: I was trying to fix this for days. It was a setting on my monitor to auto detect inputs, for some reason having this on caused an issue with the 3080, when it didn't on the Vega 64]
 
Last edited:
Soldato
Joined
12 Jan 2009
Posts
2,572
I undervolted my 3080 until my waterblock arrives. currently running it 1865mhz with 850mv and temps are around 63C max and rather than using 306w playing red dead 2 it's now using 225w.

However for some reason my 5900x is all core boosting to over 4850mhz for over 10 sec or more causing the game to crash now and again. When I had the GTX 1080 it would usually have 2 or 3 cores boost to 4950mhz for a few sec. I'll probably do a bios reset later.
I don't have an FE but undervolting is a must for a 3080. I don't know if I got a bad card but before I was hitting 78c on CP with gpu fans at 100%, sounded like a hoover. I set 881mv and I've gone from 1845-1875mhz to 1905-1920 stable at 69c after 15 mins playing the game with fans at 85% which is much quieter plus power consumption went from 340w down to 300w.
 
Soldato
Joined
18 Oct 2002
Posts
3,173
Location
Elsewhere
I've set it to 1900Mhz, GPU-Z reports 1890MHz. I wanted to try lower volts but can't seem to set it lower than 0.9v in MSI Afterburner, maybe if I reduced the clock speed. Standard clock is 2100MHz I think.

Superposition benchmark at 1440p Extreme/High/everything enabled: Score 4932 FPS: Min 29.52, Avg 36.89 w/3600. Others overclocking the card can get up to 5452, Avg 40fps.
So it’s probably a slightly lower score than stock I guess?

I have mine set to +100 core and +1000 memory and it gets to 65c max so I’m not sure if I’d see any benefit undervolting? I’ve done quite a bit of overclocking in the past but never undervolted.

The card can only boost in 15 MHz intervals which is why you’re seeing 1890 in GPU-Z.
 
Associate
Joined
23 Aug 2005
Posts
1,273
So it’s probably a slightly lower score than stock I guess?

I have mine set to +100 core and +1000 memory and it gets to 65c max so I’m not sure if I’d see any benefit undervolting? I’ve done quite a bit of overclocking in the past but never undervolted.

The card can only boost in 15 MHz intervals which is why you’re seeing 1890 in GPU-Z.
Good to know thanks. Undervolting is mainly for less power, so lower temps, fans and noise. Yeah it'll be 5-10% lower performance doing this.

Power limit and temp limit are linked so I upped those to max 109%/90C and it no longer throttles even is it reaches 85C which is the max temp in Quake 2 RTX and Superposition. Interesting that nVidia don't seem to have VRM temp or VRAM temp sensors compared to AMD, so I'm just going off the GPU temp reading.

I did up my memory by +750, this is trying to make up the lost performance from the undervolt, it might go to +1000 I haven't tried yet.

When in Control RTX all on full/max (ignoring the low fps) the GPU is at 99% load, temps reach 80C. When I switch to DLSS render at half native res, the GPU is at 80-95% load and reaches around 74C.

All these temps are relative to my low fan rpm both on GPU and my case from a quiet-as-possible-pov. And the other thing is because the FE is a lot of metal it takes a while for it to heat soak and the temps creep up, which makes testing a bit longer! :)
 
Last edited:
Soldato
Joined
18 Oct 2002
Posts
3,173
Location
Elsewhere
With my overclock on and 109% power limit I’m hitting 75 degrees in Control with everything maxxed and no DLSS, so I think I’ll just leave it as it is. I have a Fractal Meshify C which has very good airflow and one 140mm fan blowing into and below the GPU which must help quite a bit.
 
Caporegime
Joined
8 Jan 2004
Posts
32,037
Location
Rutland
Good to know thanks. Undervolting is mainly for less power, so lower temps, fans and noise. Yeah it'll be 5-10% lower performance doing this.

Power limit and temp limit are linked so I upped those to max 109%/90C and it no longer throttles even is it reaches 85C which is the max temp in Quake 2 RTX and Superposition. Interesting that nVidia don't seem to have VRM temp or VRAM temp sensors compared to AMD, so I'm just going off the GPU temp reading.

I did up my memory by +750, this is trying to make up the lost performance from the undervolt, it might go to +1000 I haven't tried yet.

When in Control RTX all on full/max (ignoring the low fps) the GPU is at 99% load, temps reach 80C. When I switch to DLSS render at half native res, the GPU is at 80-95% load and reaches around 74C.

All these temps are relative to my low fan rpm both on GPU and my case from a quiet-as-possible-pov. And the other thing is because the FE is a lot of metal it takes a while for it to heat soak and the temps creep up, which makes testing a bit longer! :)

Undervolting is useful if you're regularly power limited and you can actually achieve better than a straight max power limit overclock sometimes.

The reason is the stock voltage/frequency curve can be too generous. So a card may want to give 1.1V trying for a set clock, but at that voltage run into power limits (you can see this in afterburner) or temps can limit boost clocks too.

An undervolt will reduce the voltage delivered for each step in the voltage/frequency so you can achieve highers clocks in some cases as the power limit is less of an issue. Temps are lower to which can allow higher or more stable boost clocks.

Like you said, on the 3070FE the max power limit you can set is 109% and its really easy to hit that so an undervolt can be really helpful to maintain maximum boost clocks.
 
Last edited:
Soldato
Joined
9 Mar 2015
Posts
4,550
Location
Earth
So it’s probably a slightly lower score than stock I guess?

I have mine set to +100 core and +1000 memory and it gets to 65c max so I’m not sure if I’d see any benefit undervolting? I’ve done quite a bit of overclocking in the past but never undervolted.

The card can only boost in 15 MHz intervals which is why you’re seeing 1890 in GPU-Z.

You can do both undervolting and overclocking at the same time to get good numbers. Minstadave done a good explanation. With the generous and stock voltage curve you may be smashing against the voltage and power limit when increasing the clock speed. However dial in a lower voltage for the same clock speed and you will see the GPU not smashing against the voltage and power ceiling often.

On my 3090FE for example, stock voltage will hit around 1.0750v -1.08v or there abouts and clock to just under 2 GHz. However my day to day profile I have dialed in is 2.1Ghz but with a slightly undervolt of 1.025v though it boosts to 2130 Mhz. In actual use this settles in around 310 - 320w power usage in 4k with something like AC valhalla, more in likes of RT cyberpunk. A second profile is 0.95v for 2040 Mhz (boosts to 2055 Mhz) which sips around 290w in use. That also shows how much scaling in power is needed once you start going past 2 Ghz or so with Ampere.

So you can actually do a min/max in terms of undervolting and overclocking with these cards pretty well. By reducing voltage into the core to what's actually needed rather then a heavy handed voltage curve applied by default, your able to reduce power usage which in turn reduces heat, which can actually help boost clocks a step or two higher. Every chip is going to be different, but well worth playing around with undervolting and overclocking IMO.

Control is a good game to test this on actually that along with COD blackops I use as my sanity check more so then 3D mark which can pass fine with undervolting / overclocking while actual games like control with RT can cause things to fall over.
 
Associate
Joined
23 Aug 2005
Posts
1,273
Furmark 79C. Quake 2 RTX 88C at same fan speed. I might need to up the speeds a bit :) I always thought furmark was the brutal test? Unless I need to enable that mode.

I could just not worry about it and label all the hot stuff as 'stress testing' and gaming is running around 80C, same as Furmark.

Quake 2 RTX The Edge map at night, lit by a BFG green is pretty cool :) and flare gun makes it look like it's by camp fire. Highly recommended :)
 
Last edited:
Soldato
Joined
9 Mar 2015
Posts
4,550
Location
Earth
Furmark 79C. Quake 2 RTX 88C at same fan speed. I might need to up the speeds a bit :) I always thought furmark was the brutal test? Unless I need to enable that mode.

I could just not worry about it and label all the hot stuff as 'stress testing' and gaming is running around 80C, same as Furmark.

Quake 2 RTX The Edge map at night, lit by a BFG green is pretty cool :) and flare gun makes it look like it's by camp fire. Highly recommended :)

May be off base but may be down to Tensor and RT cores being engaged with Quake RTX while furmark does not target those bits AFAIK. Find games which engage those portions namely Cyberpunk 2077, Control etc, the power consumption vs other games is noticeably higher.
 
Soldato
Joined
18 Oct 2002
Posts
3,173
Location
Elsewhere
You can do both undervolting and overclocking at the same time to get good numbers. Minstadave done a good explanation. With the generous and stock voltage curve you may be smashing against the voltage and power limit when increasing the clock speed. However dial in a lower voltage for the same clock speed and you will see the GPU not smashing against the voltage and power ceiling often.

On my 3090FE for example, stock voltage will hit around 1.0750v -1.08v or there abouts and clock to just under 2 GHz. However my day to day profile I have dialed in is 2.1Ghz but with a slightly undervolt of 1.025v though it boosts to 2130 Mhz. In actual use this settles in around 310 - 320w power usage in 4k with something like AC valhalla, more in likes of RT cyberpunk. A second profile is 0.95v for 2040 Mhz (boosts to 2055 Mhz) which sips around 290w in use. That also shows how much scaling in power is needed once you start going past 2 Ghz or so with Ampere.

So you can actually do a min/max in terms of undervolting and overclocking with these cards pretty well. By reducing voltage into the core to what's actually needed rather then a heavy handed voltage curve applied by default, your able to reduce power usage which in turn reduces heat, which can actually help boost clocks a step or two higher. Every chip is going to be different, but well worth playing around with undervolting and overclocking IMO.

Control is a good game to test this on actually that along with COD blackops I use as my sanity check more so then 3D mark which can pass fine with undervolting / overclocking while actual games like control with RT can cause things to fall over.
I did give this a go earlier, essentially flattening off at 2070 MHz at a lower voltage than it was with my +100 core clock but unfortunately it’s not stable. I might play around with moving it up a little more again. I didn’t want to drop the clocks too much really.
 
Soldato
Joined
9 Mar 2015
Posts
4,550
Location
Earth
I did give this a go earlier, essentially flattening off at 2070 MHz at a lower voltage than it was with my +100 core clock but unfortunately it’s not stable. I might play around with moving it up a little more again. I didn’t want to drop the clocks too much really.

Going to vary from sample to sample. You will not see much of a performance difference between say 2040 MHz vs 2070 MHz but former will require a decent amount less voltage.
 
Soldato
Joined
18 Oct 2002
Posts
3,173
Location
Elsewhere
Going to vary from sample to sample. You will not see much of a performance difference between say 2040 MHz vs 2070 MHz but former will require a decent amount less voltage.
If that gives me an overall increase in FPS then I don’t mind a bit more voltage as the temps seem fine really. I use a custom fan speed curve that is pretty aggressive as I always game with headphones on too. I’ll have a bit more of a play tomorrow, thanks for the advice.
 
Associate
Joined
18 Dec 2017
Posts
338
My 3080FE arrived today, bought on a whim as I already behave an MSI 3070.

I hope I've got a decent one and the temps can be kept low, the 3070 runs at 62c on full load with barely audible fans. I have an airflow focused case (Lian Li lancool mesh 2) so hopefully I can keep it under 70c
 
Back
Top Bottom