RTX 3080 undervolting adventures

[wolf]

It looks like Ampere will make a great candidate for undervolting, because Nvidia had to push the voltages because some of the chips are not stable enough otherwise.
@wolf what software did you use? For Radeon users the driver is already awesome, I wonder if Nvidia will make their own undervolting utility...

MSI Afterburner for years now, it's very comprehensive. Would also be good if Nvidia made an internal one I agree.

don't see any reason to undervolt so far.
my 3080 runs at around 2070 Mhz while gaming and when i undervolt it runs at around 1995 Mhz at 950mv
i save like 35W under load and lose around 10 FPS.
why did i bought a 3080 in the first place then?

It's still a gratuitous upgrade for me, losing maybe a few % from stock or maybe 10% tops from an oc is still amazing when its cool and quiet, it meets or exceeds my performance target while doing so

Yeah I thought so at first too but there is a problem undervolting combine with FPS limit: let say there are situations where the FPS fall below the limit by 5-10fps, having another 100mhz core clocks at the disposal could have help maintain the FPS at the set limit.
So it's either stock voltage + overclock and set FPS limit in competitive games (PUBG, Warzone, Apex, Fortnite, etc...)
Or Undervolting + Overclock and no FPS limit in casual single player game

I feel like when we're talking like 5% clock speed it barely matters, the benefits of undervolting outweigh the performance gain. Just make sure the game is set to prefer maximum performance in NV CP, if it follows the default optimum power setting it will downclock and the effect on the fps is more pronounced.

both is completely useless..
pseudo saving a few watts to "feel better" for what? nothing.
overclock, enjoy squeezing every last frame out of it and call it a day.

Why not both? Performance is still better than stock.

Having high FPS is pointless if the game is stuttering, 95th percentile FPS FTW baby.

This too. average fps < 1% lows

On Ampere with it`s limited OC-potential the best solution seems to be the OC+UV at the highest potential that does not pwr-throttle. This will improve perf vs stock and usually yield better results that regular OC with no voltage adjustments as the lower voltage allows you to reach higher clocks. You get the best of both worlds It don`t see the problem of using a fps-limiter anyways. Not using fps-limiter will increase inputlag when you get past monitors refresh unless you use fastsync. Also not using fps-limiter will increase temps in many games and hence get to the thresholds where Ampere reduced clockspeed due to temp more easily.

All of this QFT.

 

[Deleted member 193596]

Having high FPS is pointless if the game is stuttering, 95th percentile FPS FTW baby.

well then don't use a decade old i5.

my 3080 is pinned at a 100% all the time with buttery smooth gameplay.

 

[nguyen]

well then don't use a decade old i5.
my 3080 is pinned at a 100% all the time with buttery smooth gameplay.

Smooth is subjective, can your 3080 do this

I scored 1 in Time Spy Stress Test

Intel Core i7-8700K Processor, NVIDIA GeForce RTX 2080 Ti x 1, 32768 MB, 64-bit Windows 10}

www.3dmark.com

 

[phanbuey]

power limiter on the cards is by far the biggest performance bottleneck. If you watch your "squeezing every last frame out of it" tactic you will see your clocks and frames drop well below what it would have been had you not hit the limiter.

i.e. at full blast on my 2080ti with max power headroom and a +200 OC the power limiter would knock me down from 2025 to 1770 mhz in time spy gfx test 2, and ALL THE TIME in metro exodus, in fact i would averaged 1850Mhz over a 30 min session. With an undervolt at .925mv I stay at 1950-1925 mhz all day. You get higher average clocks, you get higher average frames. Undervolt + OC is the way to go (as has been stated already by several in this thread).

 

[Vayra86]

or just place an FPS limit: less power consumptions, less input delay, better frametime consistency.
now I think the FPS limiter is a better solution overall compare to undervolting.

If you have a large excess in FPS, Fast Sync can also be a good option. Doesn't work equally well everywhere, but when it does, yummy.

 

[nguyen]

If you have a large excess in FPS, Fast Sync can also be a good option. Doesn't work equally well everywhere, but when it does, yummy.

I only tried fast sync a long time ago with Overwatch and it wasn't good, there are skipped frames which reduce motion fluidity and makes it harder to aim. Probably because my FPS weren't high enough I guess (FPS too close to screen refresh).

Now I'm looking to test out Nvidia Reflex in COD Warzone and oh boy new update is 64GB, ah hell.

 

[Vayra86]

I only tried fast sync a long time ago with Overwatch and it wasn't good, there are skipped frames which reduce motion fluidity and makes it harder to aim. Probably because my FPS weren't high enough I guess (FPS too close to screen refresh).

Now I'm looking to test out Nvidia Reflex in COD Warzone and oh boy new update is 64GB, ah hell.

Yes it gets good if you can double FPS over your refresh and I recognize what you're saying wrt Overwatch.

 

[Mussels]

Just got told the stock situation is so bad in aus i'll likely get my PRE ORDER 3080 in mid NOVEMBER

fffuuuuu AMD better launch some good cards so i can cancel and go team red if this continues

 

[BoboOOZ]

Just got told the stock situation is so bad in aus i'll likely get my PRE ORDER 3080 in mid NOVEMBER

fffuuuuu AMD better launch some good cards so i can cancel and go team red if this continues

Cough, paper launch, cough...

 

[nguyen]

Just got told the stock situation is so bad in aus i'll likely get my PRE ORDER 3080 in mid NOVEMBER

fffuuuuu AMD better launch some good cards so i can cancel and go team red if this continues

mid Nov is right when Cyberpunk 2077 release, right on time then

 

[wolf]

Cough, paper launch, cough...

More like bots and scalpers, there are plenty out in the wild now with stock rolling in in decent quantities, just go look at r/nvidia. The store I bought from at least has a good setup, they'll just let you order it and have a spot in the queue, I just happened to be spot #1 for an Asus TUF.

 

[ewitte]

I'm really sensitive to noise levels I was messing around with just lowering power targets but this way I've been able to keep near the same performance on my FTW3. Pretty much 1920Mhz constant with a few drops to 1905 at about 72C and I can barely make out the fans. The UPS says I'm usually under 450W on a water-cooled stock 3900x. I've pushed it to over 600W when I was testing with the 450W bios maxed out.

 

[Mussels]

More like bots and scalpers, there are plenty out in the wild now with stock rolling in in decent quantities, just go look at r/nvidia. The store I bought from at least has a good setup, they'll just let you order it and have a spot in the queue, I just happened to be spot #1 for an Asus TUF.

only for the USA, most other countries have F all

 

[Robert Bourgoin]

I tried undervolting my new 3800 because I was trying to get it stable , Kept crashing in stress test and games.
I purchased in with a CyberPower setup and used the guide.
https://factschronicle.com/how-to-u...ce-rtx-3080-for-better-performance-24834.html
It did not help me at all and sending the whole computer back, Going AMD when the RMA goes through. I had it with Nvidia.

 

[wolf]

sending the whole computer back, Going AMD when the RMA goes through. I had it with Nvidia

Assuming you're on the latest drivers/bios for the card, the unfortunate reality is a very slim number of parts are defective or DOA right off the manufacturers floor, you got unlucky. Equally, it's entirely possible and has been the case with AMD products too, it really just is bad luck, assuming theres no widespread issue affecting batches of products etc, which doesn't seem to be the case with 3080's.

You could have them test and replace the faulty part assuming it's all under warranty?

 

[nguyen]

I tried undervolting my new 3800 because I was trying to get it stable , Kept crashing in stress test and games.
I purchased in with a CyberPower setup and used the guide.
https://factschronicle.com/how-to-u...ce-rtx-3080-for-better-performance-24834.html
It did not help me at all and sending the whole computer back, Going AMD when the RMA goes through. I had it with Nvidia.

Have you tried taking out the RAM and pop them back in ? some parts might become loose during shipping, especially the RAM.

 

[Robert Bourgoin]

Yes I tried switching it too. I had purchased a 32 GB kit on the QVC approved list, both sets of memory same results.

I thought about just sending the video card back for exchange but what if that is not the real problem? I think it's best to send it all back.
I even tried a very high quality power supply as the one received didn't even have enough power plugs to fill the hungry board.
Tried both power supplies, latest bios on the board, switched out the video card to my older MSI 1080 and it passed all benching better than the 3080 could.
I exhausted all my resources to get it to be stable. Throwing in the towel, have to wait for RMA, I have to pay return shipping because I just want to be rid of the problematic system.
If it starts out this bad I'm sure going to be unhappy with it.
Got my eyes on AMD setup when I get my RMA first. AMD used to be the less expensive option back in the day, been a long time, times have changed price is wayyy up there.

 

[dc443]

I finally got my own 3080 (Asus TUF) this week. I'm curious for those of you who have experimented with undervolting and experiencing good performance scaling, I see Time Spy Extreme and Quake II RTX used as loads (where it's reported they run into the power limit at say 1710Mhz, my cursory testing had it hover around 1770 or 1740, though it could well drop further as the chip heats up), well my question is why not test with furmark, that generally seems to slam into the power wall in the 1500mhz range for me!

 

[Mussels]

I finally got my own 3080 (Asus TUF) this week. I'm curious for those of you who have experimented with undervolting and experiencing good performance scaling, I see Time Spy Extreme and Quake II RTX used as loads (where it's reported they run into the power limit at say 1710Mhz, my cursory testing had it hover around 1770 or 1740, though it could well drop further as the chip heats up), well my question is why not test with furmark, that generally seems to slam into the power wall in the 1500mhz range for me!

because absolutely nothing else uses power the same way furmark does, so its a useless comparison to any real world usage

 

[dc443]

because absolutely nothing else uses power the same way furmark does, so its a useless comparison to any real world usage

This is true. Anyway, with some tuning of the voltage curve I was able to get furmark to deliver almost 310fps (at max power) rather than the stock configuration delivering 270fps. It gets close to reaching 1800mhz, which it never would have done at stock. Not too shabby at all. I'll want to take a look at 3dmark and am already looking at evaluating frame time captures with the Quake timedemos to prove to myself that the adjusted curve is an improvement. I'm basically changing the curve, it's not a straight undervolt.

It seems quite doable to achieve a voltage clock curve that can produce better performance, AND reduced power consumption, AND without pushing too much toward instability. There's some headroom yet on these cards.

 

[THU31]

I am seeing some crazy power draws on my MSI 3080 Gaming X. At least in the monitoring software, however reliable that is.

First thing I did was set a flat curve of 1905 MHz at 0.9 V. And on this setting I was seeing as much as 350 W in GTA V when GPU usage was close to max. Way over 300 W in other games as well (and TessMark, which I like using for testing stability). With a custom fan profile I saw a max temperature of 78* at ~1750 RPM, which is quite high.

Then I tried 1800 MHz at 0.8 V and I am still seeing just over 250 W in extreme cases. Temps and fan speeds are way lower, though.

Seems crazy to me that you could see such high power draw at such low voltages and clocks.

 

[Mussels]

I am seeing some crazy power draws on my MSI 3080 Gaming X. At least in the monitoring software, however reliable that is.

First thing I did was set a flat curve of 1905 MHz at 0.9 V. And on this setting I was seeing as much as 350 W in GTA V when GPU usage was close to max. Way over 300 W in other games as well (and TessMark, which I like using for testing stability). With a custom fan profile I saw a max temperature of 78* at ~1750 RPM, which is quite high.

Then I tried 1800 MHz at 0.8 V and I am still seeing just over 250 W in extreme cases. Temps and fan speeds are way lower, though.

Seems crazy to me that you could see such high power draw at such low voltages and clocks.

cause its a lot of little chips in parallel, running at low clocks and low volts

 

[THU31]

I would think the Samsung process would be very efficient at low clocks and low voltages, but I guess it is just terrible all around. I am still very happy with the card, but I expected a lot more from undervolting.

I ran my 2070 SUPER at 1905 MHz and 0.95 V and it was way below its TDP. The 3080 is up to twice as fast at this setting, but it consumes twice as much power. So there is no efficiency gained from either the new process or the new architecture.

Remember Maxwell, which had double the efficiency on the same process? That was insane.

I guess they knew what AMD had with Big Navi, so they had to rush this entire generation, which explains both the lack of efficiency and the lack of supply.

 

[nguyen]

I would think the Samsung process would be very efficient at low clocks and low voltages, but I guess it is just terrible all around. I am still very happy with the card, but I expected a lot more from undervolting.

I ran my 2070 SUPER at 1905 MHz and 0.95 V and it was way below its TDP. The 3080 is up to twice as fast at this setting, but it consumes twice as much power. So there is no efficiency gained from either the new process or the new architecture.

Remember Maxwell, which had double the efficiency on the same process? That was insane.

I guess they knew what AMD had with Big Navi, so they had to rush this entire generation, which explains both the lack of efficiency and the lack of supply.

it's game dependant, with AC Valhalla my 3090 consume only 220W at 1860mhz/800mV and CP2077 at the same undervolt use 290W. AC Valhalla looks like shit next to CP2077 (RTX+DLSS ON) yet they both run around 60fps with 3440x1440 Ultra settings on my 3090.

You can always drag the power limit slider to what you want and let GPU boost does its thing.

 

[THU31]

Do you have any idea why my voltage curve keeps changing slightly every time I restart my computer?

For example, I set a flat line at 1800 MHz and 0.8 V, but when I restart my PC it will either go up or down by 15-30 MHz. I have always had this issue, on my 1080 and 2070 as well. It is kind of annoying.

Sometimes it will even change to a weird curve.

Proper:


After restart: