3080 Undervolting

[Psycho Sonny]

Playing around with some undervolting on my 3080 Ti FE.
I've set it to 1935 @887mV.

Seems to peak at 81°C in Timespy.
GTA V peaks at 70°C with a peak clock of 1960.
Seems a bit quieter overall as well.

Very new to this, don't know if any good.

The sweet spot from the video is

"1900 at 887mV is the sweet spot"

You want to stick with that voltage for sure. You are getting 35 higher than what was recommended in the video and a few people on here have apparently been able to un games ta but it would crash randomly. I would be careful unless of course you hit the silicone lottery. You are running at what I would consider very very good settings and any more gains from here on in would be minimal and likley come at higher core voltage causing temps and noise to increase.

 

[Skynet5]

The test is defo quake 2 rtx for 30mins.

Lol

 

[Dicehunter]

it really does clock was in 1700mhz - high 1600mhz range RTX doesnt add anything to the game still looks crap lol need to find something else

Control makes a butt ton of use of RT, Could find a repeatable section to walk through with a lot of lights and reflections, Don't turn on DLSS and see how it goes.

 

[Jae-So]

The sweet spot from the video is

"1900 at 887mV is the sweet spot"

You want to stick with that voltage for sure. You are getting 35 higher than what was recommended in the video and a few people on here have apparently been able to un games ta but it would crash randomly. I would be careful unless of course you hit the silicone lottery. You are running at what I would consider very very good settings and any more gains from here on in would be minimal and likley come at higher core voltage causing temps and noise to increase.

Thanks makes sense
Think I'll drop it down to 1900, as you say gains are minimal.

 

[uscool]

Control makes a butt ton of use of RT, Could find a repeatable section to walk through with a lot of lights and reflections, Don't turn on DLSS and see how it goes.

Tried Control still dont think it looks all that impressive wasnt blown away but some parts did look quite good , at stock the clock was around 1800mhz looking at youtube with same game and FE card can see people getting higher stock clocks my temps seem very good so cant be that guess its lottery with each chip also heard newer drivers nvidia reduced the boost clocks , with undervolt can sustain 1905 with odd jumps to 1920 and it seemed stable the 30mins I played

 

[Jae-So]

1905Mhz @ 887mV - RTX3080 Ti FE.
Very little performance loss if any.
Need a CPU update now.

 

[Dicehunter]

Tried Control still dont think it looks all that impressive wasnt blown away but some parts did look quite good , at stock the clock was around 1800mhz looking at youtube with same game and FE card can see people getting higher stock clocks my temps seem very good so cant be that guess its lottery with each chip also heard newer drivers nvidia reduced the boost clocks , with undervolt can sustain 1905 with odd jumps to 1920 and it seemed stable the 30mins I played

Could always try Metro Exodus Enhanced Edition, RT global illumination, RT emissive lighting, RT reflections, Quite a workout for a GPU

 

[Rob43]

Could always try Metro Exodus Enhanced Edition, RT global illumination, RT emissive lighting, RT reflections, Quite a workout for a GPU

This game will definitely let you know if your undervolt is 100% stable. I've been running 875mv @1890mhz on a 3080FE with no issues with any games.
Fired up exodus ee & couldn't play for more than 5mins without it locking up. Turns out was my undervolt, set gpu all back to stock & now its running with no issues.

Rob

 

[Rob43]

This game will definitely let you know if your undervolt is 100% stable. I've been running 875mv @1890mhz on a 3080FE with no issues with any games.
Fired up exodus ee & couldn't play for more than 5mins without it locking up. Turns out was my undervolt, set gpu all back to stock & now its running with no issues.

Rob

 

[Psycho Sonny]

1905Mhz @ 887mV - RTX3080 Ti FE.
Very little performance loss if any.
Need a CPU update now.

I score about 1,000 points higher iirc with a 5600X

 

[Jae-So]

I score about 1,000 points higher iirc with a 5600X

Total score or GPU score?
I fully expect that on the total score.

Main thing I was trying to get across was I didn't see a drop off on my previous score (could've added that picture) after the undervolt.

 

[Psycho Sonny]

Total score or GPU score?
I fully expect that on the total score.

Main thing I was trying to get across was I didn't see a drop off on my previous score (could've added that picture) after the undervolt.

I just checked an no your score is very close to mine. For some reason I thought I got 14K+

https://www.3dmark.com/pr/1224991

I got 100 points more.

Graphics Score
13474

I actually used your settings for it lol. 1935 at 887mv

I also ran heaven benchmark after and it got through it fine. So to me it's "stable enough".

I might drop it to 1900 as I can see it throttles down to that anyway.

 

[Jae-So]

I just checked an no your score is very close to mine. For some reason I thought I got 14K+

https://www.3dmark.com/pr/1224991

I got 100 points more.

Graphics Score
13474

I actually used your settings for it lol. 1935 at 887mv

I also ran heaven benchmark after and it got through it fine. So to me it's "stable enough".

I might drop it to 1900 as I can see it throttles down to that anyway.

Nice that gives me some hope that the CPU still has enough oomph to at least let the GPU do its thing at high-Res.
I'm actually playing on my 4K TV more than my UW 144Hz screen at the moment so should minimise any bottleneck.

I've dropped down to 1900. In benchmarks I've found it actually does slighty better than 1935, but the difference is all within the margin of error.

 

[Psycho Sonny]

Nice that gives me some hope that the CPU still has enough oomph to at least let the GPU do its thing at high-Res.
I'm actually playing on my 4K TV more than my UW 144Hz screen at the moment so should minimise any bottleneck.

I've dropped down to 1900. In benchmarks I've found it actually does slighty better than 1935, but the difference is all within the margin of error.

Yeah I'll drop mine down to 1900 tomorrow. Your CPU is fine if it's only 100 points off mine. I'm also using PBO to overclock my CPU and I have 32GB of 3600mhz cl 16 ram. 4*8GB sticks to get the benefit of dual rank.

So I'm pretty much maxed out without getting ridiculous in terms of law of diminishing returns and majority of the build has been focused on value for money apart from the RGB stuff.

You shouldn't need to upgrade that CPU for at least 5+ years I would think. I don't plan on upgrading mine until 2 more generations have passed and ddr5 is available for peanuts and the timings and speeds have reached max potential.

It's not like anything I have currently will hold me back. My gaming pc is at the point where there is literally nothing else left to do or upgrade. Recently got another TB of SSD so now have 3TB total.

Hopefully they do a 5XXX series refresh like they did with the 3600XT. It means I could get another upgrade yet without ditching DDR4 or a new mobo.


Just realised. I ran port royal benchmark not timespy.

Port royal is GPU focused only and really pushes the GPU.

Could you run port royal? It takes CPU out of the equation. I'll need to run timespy to see if my CPU is better than yours and subsequently the score.

 

[Jae-So]

Yeah I'll drop mine down to 1900 tomorrow. Your CPU is fine if it's only 100 points off mine. I'm also using PBO to overclock my CPU and I have 32GB of 3600mhz cl 16 ram. 4*8GB sticks to get the benefit of dual rank.

So I'm pretty much maxed out without getting ridiculous in terms of law of diminishing returns and majority of the build has been focused on value for money apart from the RGB stuff.

You shouldn't need to upgrade that CPU for at least 5+ years I would think. I don't plan on upgrading mine until 2 more generations have passed and ddr5 is available for peanuts and the timings and speeds have reached max potential.

It's not like anything I have currently will hold me back. My gaming pc is at the point where there is literally nothing else left to do or upgrade. Recently got another TB of SSD so now have 3TB total.

Hopefully they do a 5XXX series refresh like they did with the 3600XT. It means I could get another upgrade yet without ditching DDR4 or a new mobo.


Just realised. I ran port royal benchmark not timespy.

Port royal is GPU focused only and really pushes the GPU.

Could you run port royal? It takes CPU out of the equation. I'll need to run timespy to see if my CPU is better than yours and subsequently the score.

Yeh I'll run Port Royal tomorrow/later and report.

I'm glad you said about my CPU, most think I need to upgrade now, but I don't see the point until I can get a new platform with DDR5 support. I'm thinking more like 3 years though.

I have a tonne of SSDs myself too lol.
About 4TB of SATA SSDs and 500MB NVMe
I haven't bothered putting in a large NVMe yet as I really need a new platform to take advantage of Gen.4 drives.

 

[Jae-So]

Y

Just realised. I ran port royal benchmark not timespy.

Port royal is GPU focused only and really pushes the GPU.

Could you run port royal? It takes CPU out of the equation. I'll need to run timespy to see if my CPU is better than yours and subsequently the score.

Just remembered I ran Port Royal previously.
I'm dropping about 700 points (6% or so) to you.

 

[uscool]

so far settled for 1905@ 893mV and boosts to 1920 seems the sweet spot, but I decided to try 1950mhz @950mV only to see if bench scores improved but they got worse ?? even though the temps only reached 69c the clocks started to fluctuate all over 1875-1965 didnt think temps of only 69c would make them fluctuate that much ?

 

[Jae-So]

so far settled for 1905@ 893mV and boosts to 1920 seems the sweet spot, but I decided to try 1950mhz @950mV only to see if bench scores improved but they got worse ?? even though the temps only reached 69c the clocks started to fluctuate all over 1875-1965 didnt think temps of only 69c would make them fluctuate that much ?

Yeh it's weird, setting higher clocks was counterproductive for me too. That's why I settled at 1905 @ 887mV.

 

[Psycho Sonny]

Just remembered I ran Port Royal previously.
I'm dropping about 700 points (6% or so) to you.

Okay I put clocks down to 1908 and it seems to stick solidly at 1920 on boost. 3080ti has better silicone than 3080 so this seems like the sweet spot for my card.

I ran timespy and got 16612. With this undervolt my card is pulling 330W-350W during benches I don't know what the 3080ti pulls at stock but 3080 pulled 370w and pulled 80W less when undervolted at similar settings.

https://www.3dmark.com/3dm/66247955?

CPU score is circa 8800, GPU score circa 19700. Looks like your CPU is indeed holding you back slightly. Not enough to upgrade yet IMO.

https://www.3dmark.com/compare/spy/22596400/spy/22960133

3080ti is 9% faster than 3080 in Time Spy my old card vs new. My 3080 was scoring same as your 3080ti does currently. So looks like you are being held back roughly 8-9% on CPU limited games. But on non cpu limited games the score is far closer with my 3080ti score so not a huge issue IMO.

 

[Jae-So]

Okay I put clocks down to 1908 and it seems to stick solidly at 1920 on boost. 3080ti has better silicone than 3080 so this seems like the sweet spot for my card.

I ran timespy and got 16612. With this undervolt my card is pulling 330W-350W during benches I don't know what the 3080ti pulls at stock but 3080 pulled 370w and pulled 80W less when undervolted at similar settings.

https://www.3dmark.com/3dm/66247955?

CPU score is circa 8800, GPU score circa 19700. Looks like your CPU is indeed holding you back slightly. Not enough to upgrade yet IMO.

https://www.3dmark.com/compare/spy/22596400/spy/22960133

3080ti is 9% faster than 3080 in Time Spy my old card vs new. My 3080 was scoring same as your 3080ti does currently. So looks like you are being held back roughly 8-9% on CPU limited games. But on non cpu limited games the score is far closer with my 3080ti score so not a huge issue IMO.

Awesome, really appreciate you running this for my context.

Yeh looks like I am leaving some performance on the table but yeh I don't think it's something I need to address at the moment - I can live with a 10% bottleneck.

Looks like I'm set for a couple of years
Was playing some games on my 4K TV last night and it all just looked and ran great.

3080 Ti is a great card to have and maybe I'll finally dive back into Cyberpunk 2077, RT enabled.