A BIG Disclaimer I forgot to put in the video: Just like with overclocking you must stress test your undervolt for stability. Use a bunch of games, and some 3D synthetic stress tests like 3DMARK timespy to ensure you don't suffer from any crashes.
Undervolting IS still overclocking so stability has to be checked for sure. When "undervolting", you're overclocking. For example, if stock you got 2480 MHz at 0.875v, the undervolt will give you 2700 MHz at that same 0.875v so you're asking for more frequency relative to voltage than stock so it's overclocking. I feel like a lot of people don't realise this. Sure, the card's max frequency hasn't increased (eg. getting it to 3500 MHz) but the frequency has still increased for the amount of voltage so it's overclocking.
@@samburti8587 You just ask less voltage for the same frequency. For exemple, stock I'm at 1.025v for 2700 mhz. I set 2700 mhz to .950v and flat the curve... So if the card got enough voltage to get 2700 with that voltage, after that I will be fine. I don't ask more frequency for the same voltage, or give more voltage. The card is fine to run 2700 and fine to receive .950, it's just to know if .950 is enough for that speed... Yeah of course it need test, but it's not overclocking.
The manufacturers (Nvidia, AMD, intel, for both cpu's and gpu's) have blasted all the new products with brute force levels of power designed to increase the 'rate' at which silicon passes spec, so they can call it an increase in wafer "yield". 95-99% of the silicon in the wild doesn't need the juice which manufacturer spec calls for. And frankly, they're dancing the products near the flame of thermal failure, and decreased life-span. The _performance_ sweet spot for both my 7900X and 4090 TUF OC, has been to undervolt both. And out of interest for longevity, I tightened up the thermal limits as well.
suggestion: stop moving the bar charts around based on average fps. it makes it confusing to read the charts from one game to another. just leave it as OC, stock, 90% and 80% all the way regardless of the fps/power draw and it will be much easier to follow along.
Definitely mandatory. I gave up halfway through as I realized I had been misreading a lot of the charts already.. These are good findings, but could be presented much better and cleaner.
I just tried this, expecting a 5-10% drop in performance. Instead, I gained 2-3% in performance, while getting lower temps and lower power consumption... I get better benchmark results from 90% power limit and 0.95 undervolt and small OC on memory compared to 133% power limit and default voltage curve. Thanks for the video!
That's because you're overclocking your card. Undervolting means your card will attempt the same clocks as before but at lower voltages = overclocking. That also means your card, at the same voltages as before, will attempt higher frequencies = overclocking. Undervolting is the same as overclocking as you're asking for a higher amount of frequency relative to voltage. The 90% power limit on the other hand simply limits performance by reducing the frequency/voltage part of the curve you're using for that moment so hitting it will always slow you down no matter what.
@@samburti8587 If someone told you they were going to overclock their GPU, I think you'd have an idea of what they meant. If another person told you they were going to undervolt their GPU, I think you'd have an idea that they meant something else. You would then use this opportunity to insist that they are the same thing when, clearly, two different ideas were expressed ... and you understood that at the start. You introduce logical confusion by ignoring the idea that 'performance' can be measured based upon what the person wants to achieve. You would insist that performance can only be defined by fps. This is the tactic of a brute.
Been using similar settings myself (80% power, 900mv at 2625 Mhz, memory +1000). Feels nice to outperform my previous GPU, a 3080 Ti, so much, at lower power draw.
Awesome stuff. I have a similar system with a 5800X and 4090 FE. I've set a 70% power limit and haven't toyed with undervolting the card just yet, but I'll go ahead and give it a try. It's amazing how efficient these top end GPUs and CPUs that have recently launched can be, and how it gets overlooked because Intel/AMD/Nvidia decide to push them to the limit.
70% power limit seems to work great. I have had no issues running games at 4k 120hz, so its nice to know if I ever run into issues, I can just raise the power limits and get a little extra performance.
I'm a newbie and I'm scared to mess with the volts so I just dropped the Power limit to 60-70% and games run amazing anyways with that said do you guys recommend me trying to mess with it its alot of money I'm playing with as a new pc user
Pretty useful video but the thermal testing was next to pointless with the fans running at different speeds across the 3 modes. Should have locked the fan to a particular speed for all tests.
Another cool addition to a graph would be frame capped at 120. Just to show how much power it draws in usable situations. And maybe compare it to last gen. I guess I could do it… have a 3080 to compare with. I bet it pulls under 200w in some games. Nuts
I found that most games only pull 350w tops. So same as my 3080 10g. I thought about changing my hx750i psu for exactly 0 seconds… I knew it could be close while OCed, but also assumed it would be very efficient at 350 on Tsmc 4n. I was right. Total sys power draw is 350w in several games I play frame capped at 120 on Lg c1.
How do you hear coil whine over your headset and machine guns/explosions/swords clanging + game music? I never understood the complaining about fan noise etc, it must be as loud as a hair dryer for you to hear that over your headset.
I don't know how I have avoided coil whine my entire life. I havae gamed with 2080-3080-3090(x3 - yes 3 different cards)-4090 and I have never heard coil whine.
I did some test with my 4090 in furmark. 950mv undervolt gave me 744 fps / 355w / 65c (2800 mhz@ 0.95v +1500 ram 80 power limit) stock gave me 786 fps / 450w / 73C full blown OC gave me 796 FPS / 490w / 77C (+200 core +1500 ram 133% power limit.) so we are looking at a decrease of 6% performance for 30% less power when comparing the overclock to the undervolt. If we look at it the other way, you would need to increase undervolted card's power draw by 40% to gain 7% performance. im going to stick to my undervolt, the fans are loud as hell when it gets hot XD. i also tried just playing with the power slider. at 80% power limit the results were similar to the .950 undervolt but voltage was always 10-15w higher and performance the exact same.
I had a hard time telling WTF was going on until I realized you were moving the configs around. You should keep the places constant for something like this.
Kinda wish you would have been a little more detailed on how you do this, the very short clip shows you dragging one slider down to a random number 172, whats the significance of that? Then you drag the whole curve higher which isnt very clear how this is underclocking the card. I want to do this but am almost more confused after watching this video.
Gamers, when the GPU doesn't use the full power potential getting slightly less fps improving power draw: WE NEED TO CHANGE THAT! Gamers now, when the GPU is using its max potential from the start: WE NEED TO CHANGE THAT!
I have a SFF 850 watt coolermaster driving a 13900k power limited to 180 watts and 4090 at stock with no power limit (450 watts) System runs great. A quality 850 watt PSU is fine even at stock, as long as you don't let the CPU go bananas
I'm looking to still run all my games in max setting (1440p), but getting lower temperature. I just try undervolting at .950 @ 2700 mhz with +1000 memory clock and power limit 80% and I have pretty much the same performance than default, but -8 degree celcius with the Benchmark of Rdr2 (yes I know, I love to use in-game benchmark). I think the best, is to keep high frequency, but at the lowest voltage possible. I think .950 is really decent. During the Benchmark, the voltage was .960... I don't know why it goes up a little but, but it's still lower than the default (1.050 I think). I see a lot of people talking about just lower the power limit... I tried and it didn't work. Everything by default with the power limit at 80%, the voltage was like 90% of the time at 1.050 and just get to 1 a little bit. I had average FPS -1, so pretty much the same, but with 1 degree hotter (not a average, I just look the maximum temperature I got during the test).
I do -300 mhz on Core, then find .970v in curve optimizer and pull that up to 2820mhz and hit checkmark. Running stable for over 10 months on RTX 4090 Gigabyte Gaming OC. 2805 mhz is another great value. I wouldn't touch memory, memory overclocking on GPUs really makes cards unstable and degrades the chip too quick.
Undervolting + power limit doesn't make sense in comparison to stock for power consumption. It would be better to run both on 90% limit, or without any.
so what i am getting with my OC is hdr 4k settings in max no dlss either is not over heating but a type of lag. but my point is im getting over 120fps when i first play the game then random drops to 45fps mostly when i inventory and camp site or a sudden change of room. so cpu goes high too but gpu drops to 34%-56 % or worse. just awful. so the card easily handles any settings for awhile but then doesnt handle it. i have a 13900ks cpu too so its not going to stop it.
Yup. If I'm not mistaken, the 2nd undervolting configuration has an average loss of 2.21% FPS while decreasing power consumption by nearly 20%. That's like 30 something Watts per FPS lost.
I waterblocked my FE 4090. I just set GPU to use 600wats and overclocked with after burner till stable, uses 450watts with no fans! when running 4k/8k ray tracing. BUT - Then I limit frame rates around 200 so games like Destiny and Diablo barely tax the card/ hence I get good power to performance because it never redlines, not sure the logic works out but seems to offset cons while ensuring I can hit peak OC performance. I would not suggest doing what this guy suggests..., as I can get this card to 35fps on a bunch of games I play 4k/ray tracking and with frame gen and above tweaks I get 70-90 vs. sub 60 stock..... so barely good enough. Jk
Can you please explain why you would drop the entire curve down first then spike up a single point where you flatten, it makes no sense to me why not just leave the curve as close to default as possible for stability sake?
RTX 4090 AMP Extreme, 2.7 GHz, +1500mem, 950 mV, 90% power and Cyberpunk crashes, I believe it's not enough power. Any thoughts? Edit: After a few tests I just set it at 975 mV and it works perfect.
Great video btw! I'm trying to set the curve as you showed in the video, but when pulling the dot above the 950mv only that point goes up. It doesn't level everything out like yours did in the video. Were you holding another key when moving the point by any chance?
That's the way it is have to be. You then hit the Checkmark on main MSI Afterburner window, and it will even the rest of spots. Make sure to do -X value before doing that (on core) I do -300 mhz on Core, then find .970v in curve optimizer and pull that up to 2820mhz and hit checkmark. Running stable for over 10 months on RTX 4090 Gigabyte Gaming OC. 2805 mhz is another great value. -300 is very aggressive though -200/-250 is a safe bet I wouldn't touch memory, memory overclocking on GPUs really makes cards unstable and degrades the chip too quick.
Overclocking should be a choice, not nVidia default. Most people won't do it, and many of them because they want efficient hardware. The 4090 out of the box is anything BUT efficient. It is running way beyond the efficiency curve into power hog territory. Driver tweaking does not solve this. It does not run like that in linux, and it does not run like that if you are running from an USB stick, it does not run like that if you experience a crash, and it does not run like that if you put it into a different PC. So nVidia made overclocking the stock by pushing 150 more watts into the card without us asking for it. If there was a switch on the GPU that limited it to 300W that would be fine for me, but there isn't.
You sir, have no clue what you are talking about. The 4090 runs cooler, and draws less power than my previous 3080. I can tell, you do not own a 4090. You'd never make such a ridiculous comment. The 4090 is MORE EFFICIENT, than previous Gen. You are simply parroting all the early reports and guesses made BEFORE people actually got the card.
Spongy is correct. The 4090 draws less power than the 3090 and can be undervolted by 20% for single digit performance loss. For which I thank all the voltage outrage over the previous months for achieving. I'm betting they redesigned the card improving efficiency after the backlash resulting in a far better product.
@@geronimo5537 sorry guys but you are being delusional. The 3090 has a TBP of 350W, and won't draw what the 4090 can. You are perhaps confusing with the 3090Ti. But the 4090 is not the top model of the 40 series so you should not compare it to the 3090Ti but the 3090. But even if it was it does not matter. It is running completely off the charts in inefficiency. By limiting the card to 300W you lose ~10% performance. So barely over 10% performance for 50% more power. That's a completely overclocked card. It does not matter that it has a 3Kg cooler that keep it cool and quiet, it still is not efficient. I'm not asking for much. I'm simply asking for a toggle on the card that reverts this overclock back into an acceptable efficiency range.
@@marsovac well if you want some details. Seems the 4090 will benchmark the same as a 3090 if power is cut down to 40-50% usage. That is the same performance as a 3090 for near half the watts a 3090 needs. The 4090 also consumes less power in general compared to the previous 3090 as well. Which at this point is very well documented. Any comparison video of the two cards is going to show that information.
um i guess my 4090 xtreme airo that came from zotac is messed up cause it's BASE is the overclock at 2580 and it boosts out of the box without me touching it or the bios saying "we're boosting you" to 2980. I dont understand how this card shows in everything with the base clock as 2580 and boost 2780 then BOOSTS to 2980 (I have logs) without me changing a darn thing in afterburner or firestorm.
Have you witnessed the same behavior as Ali did from Optimum Tech? His 4090 lost performance with his UV even when the core clock was higher than stock. He blamed the decrease in "gaming clock". I'd be interested, because the way I UV is to get more performance at lower power draw and decreased thermals.
What's a good way to lower temps on this card? I have a msi liquid x and it can get insanely hot. My entire room feels like it was back in the xbox 360 era with it being a heater.
Great video and clear instructions, curious if you're going to show the smaller form factor build you mentioned in another vid or would that be a while away? Interested to see how the card fairs in those situations.
I know this is an older vdo. Maybe things have changed. I currently have my 4090 overclocked +235 core , +1317 mem , max power , the most ive seen it draw is 560watts. I'm still on my 800psu. I do agree with you that core increases barely gave me gains, I think I saw 2 fps more in Red Dead Redemption on a 120fps average scale. The memory overclock on the other hand gave me like +10 fps. I'm not sure why I'm drawing so much more wattage than you but I did see one of your benchmarks that showed the Overclock being slower than stock. This makes me question how stable your overclock is, you might actually do better with less overclock and it'll pull less wattage automatically, this would skew your results to show significant gains for the power usage. Overall though, I'm seeing about 10% gains in overclock. This amount is significant if you are looking at it versus a 3090ti or something it gained 50% performance on. In a resolution lower than 4k, 10% is a decent amount. I guess the bigger question is, other than the concern over powerbills , whats real legitimite use of undervolting. My answer would be, to be able to use the card in a PSU limited scenario. Not having to spend more money on a new PSU is worth more than the questionable utility bills savings I believe. It's small energy expenditure to be honest. I 'm old school where our lightbulbs used alot of watts. Turning off lights and appliances can make up for the power savings. Anyways rant over, I'm sure everyone has their own idea over what they want out of their card. I didn't spend mucho money to gimp its performance. I didn't get an AIB with a huge cooler problem in order to try to reduce the heat by reducing performance. The perfect spot place I would love an undervolt is if I can get the same overclock. Maybe its overshooting the required voltage at the moment.
Nice video. Why are you both using power limit and undervolt? I strongly advise against purely using power limit, as it uses more power with less demanding games versus an undervolt. But what is the benefit of using undervolt + power limit? With the undervolt it will hardly ever hit the power limit. Some more information for readers: I'm running my rtx 4090 at 900 mv with 100% power target. it is a 3% difference, and with my VRR screen, i certainly can not see the difference between 100 fps and 103 fps. With a lot of games maxed out, my RTX 4090 stays below 300 watts. I found that only with Metro exodus, the power usage can go up to 380 watts! With Forza Horizon 5 maxed out, with RT it only uses about 260 watts. This is with the 900mV undervolt.
Unfortunately the connector melting issue has little to do with power draw but more so to do with misalignment of the pins due to the adapter being bent. I'm hoping Nvidia will sent out right angle adapters to people for free because that is just unacceptable.
The real problem is the internal foil pins cracking and arcing inside the connector. Basically any horizonal stress can cause this. So its best to keep the adaptor very stress free and limited to vertical bends.
Hey do you have a video on PBO and CO overclocking the 5800X? I have one to. I usually just do an all core of 4600 to 4700 but would like to try PBO+CO
You can find that video here. It really isn't an in-depth tutorial video on how to use PBO2 and curve optimizer but I do briefly cover the methodology so that anyone can recreate the same settings themselves. ua-cam.com/video/wp9OmI7Xbnc/v-deo.html
hmm, interesting... So the 3 settings to change in MSI afterburner are Power Limit, Core Clock and Memory clock right? You don't have to change anything for "voltage".
4080 to? My self clock up to 3,1 ghz msi suprim x white very good cooling and case, 125 powerlimit and 88c monster! Hade asus tuf pc 4080, only 2,8ghz and coilwhine
I don't know if it matters, but did you have different settings for the Benchmark? Or was it somehow shots from a different time of day? The shadows look very different.
I have a Intel i9 108850 running more or less stock I just set all cores to 5Ghz as a minimum and left the regular higher speed cores as is. I have 64GB memory and 10 Corsair LS13 Led fans and li strimmer lights. I have a RMx 850watt PSU. If I undervolt and do what you have done is my 850 watt PSU enough with a 4090?
I don't know why but my msi rtx 4090 suprim only takes 280w on full load and it runs at 2800mhz don't know why and no fps drop and got 1000w psu so I'm not complaining lol
Power Limit 85/ Target Temp 88/Core Clock +205/Mem Clock +1500/ Fan Auto. Max I see for watage is 380w(down from 430+w) but the clock will just sit ther at 2800+ (I've seen it sit at 2900+ in some gaming Benchmark Tools). I'm done tinkering yay! MSi 4090+5800x3D. P.S. I think Nvidia is setting us up for a "more efficiency" scam with the 5000 series. My card should not be this Stable with a near max clock and a huge chunk of wattage missing(17 Benchmarks cleared with no crashes including the 3D Mark suite).
How does, when undervolted, the power consumption develop when you instead of introducing a powerlimit use a fps limit, like 100 or 90 fps on a monitor with g-sync? The reason I am asking is because I am puzzled which one is more efficient maybe a mix of both fares best? My target is 90 fps, but I am not sure how the results are especially because atm I only use a fps limit, I lack the equipments to test that.
I did the test with borderlands 3, on 1440p, ultra settings. Stock: Peak power consumption: 435W. 144fps limiter (via nvidia driver): Peak power consumption: 354W 144fps limit + UV (2550Mhz @0.900mV): 201W.
What was your stock Speed way score? out of the box? im getting around 9630 which i think is on the low end for the same card as you. it should be closer to 10k. Can you tell me what you were getting?
For my daily settings I will be running with the 2nd undervolted config, where I targeted 2.62GHz at 900mV, overclocked mem +1500, and limited power to 80%. I'm okay with losing a tiny smidgen of performance to save like 20-30% power and lower noise.
@@katates My card's coil whine is similar to yours, and I've seen other samples from others 4090s and it also sounds similar. I think its just going to be normal with these cards, considering how large the die is, how many components are on the PCB, etc. Undervolting does actually lower coil whine a bit because there's not as much power being shoved through those inductors and coils.
Hello i need help, when i preset 900mv 2662 mhz 1500 memory clock and i try furmark my pc crush, and my image gets black point, any help? Sorry for my english
No, they don't have to. I don't do overclocking to have less FPS and worse results in benchmarks by undervolting. If you can't afford electricity, don't buy an RTX 4090. This subject of undervolting is boring.
Why are you skeptical? The video provides benchmark results which demonstrate a 1 FPS AVG and 1 FPS 1% low difference between stock and 80% power undervolt. And if you want, you can just test it yourself.
@@Goodbutevilgenius for staters why put +1500 mem on the under volt cards a not and on the stock is he’s trying to compare like for like. This generation it’s the memory that has big impact on performance compare to the core. So the results have been skewed to show little to no loss. Not saying the difference will be big but there will be some performance drop off. So by saying no performance loss in the title thumbnail is false.
At the end of the day its up to you. I personally just felt that If I can save on power while getting the same exact experience as stock, I'll definitely make the trade. Plus I also game at a capped framerate on my LG C9 to stay within the gsync threshold.
@@DannyzReviews I’m old an lazy. I’ll mess with the curve if it becomes necessary. I used to have to do it twice almost every time I turned on my system with my 2080 Ti’s because the voltage curve was buggy with those and wouldn’t ever save correctly if I restarted. They were severely power limited and unstable with the auto curve, so it was pretty much necessary. The upside is I got really really good at tuning a custom curve 😆
A BIG Disclaimer I forgot to put in the video:
Just like with overclocking you must stress test your undervolt for stability. Use a bunch of games, and some 3D synthetic stress tests like 3DMARK timespy to ensure you don't suffer from any crashes.
So far games just don't work for more than a benchmark run on option 2.
Option 1 is perfect. Lose 10 percent power consumption, a tiny bit of heat, but keep like 98 percent of performance of stock at 950 mv.
Undervolting IS still overclocking so stability has to be checked for sure. When "undervolting", you're overclocking. For example, if stock you got 2480 MHz at 0.875v, the undervolt will give you 2700 MHz at that same 0.875v so you're asking for more frequency relative to voltage than stock so it's overclocking. I feel like a lot of people don't realise this. Sure, the card's max frequency hasn't increased (eg. getting it to 3500 MHz) but the frequency has still increased for the amount of voltage so it's overclocking.
@@samburti8587 You just ask less voltage for the same frequency. For exemple, stock I'm at 1.025v for 2700 mhz. I set 2700 mhz to .950v and flat the curve... So if the card got enough voltage to get 2700 with that voltage, after that I will be fine. I don't ask more frequency for the same voltage, or give more voltage. The card is fine to run 2700 and fine to receive .950, it's just to know if .950 is enough for that speed... Yeah of course it need test, but it's not overclocking.
The manufacturers (Nvidia, AMD, intel, for both cpu's and gpu's) have blasted all the new products with brute force levels of power designed to increase the 'rate' at which silicon passes spec, so they can call it an increase in wafer "yield". 95-99% of the silicon in the wild doesn't need the juice which manufacturer spec calls for. And frankly, they're dancing the products near the flame of thermal failure, and decreased life-span.
The _performance_ sweet spot for both my 7900X and 4090 TUF OC, has been to undervolt both. And out of interest for longevity, I tightened up the thermal limits as well.
suggestion: stop moving the bar charts around based on average fps. it makes it confusing to read the charts from one game to another. just leave it as OC, stock, 90% and 80% all the way regardless of the fps/power draw and it will be much easier to follow along.
Suggestion? MANDATORY!
Definitely mandatory. I gave up halfway through as I realized I had been misreading a lot of the charts already.. These are good findings, but could be presented much better and cleaner.
Definitely mandatory, this was REALLY irritating.
Exactly
True, seems like people dont even watch their own content its stupid
I just tried this, expecting a 5-10% drop in performance. Instead, I gained 2-3% in performance, while getting lower temps and lower power consumption... I get better benchmark results from 90% power limit and 0.95 undervolt and small OC on memory compared to 133% power limit and default voltage curve.
Thanks for the video!
That's because you're overclocking your card. Undervolting means your card will attempt the same clocks as before but at lower voltages = overclocking. That also means your card, at the same voltages as before, will attempt higher frequencies = overclocking. Undervolting is the same as overclocking as you're asking for a higher amount of frequency relative to voltage.
The 90% power limit on the other hand simply limits performance by reducing the frequency/voltage part of the curve you're using for that moment so hitting it will always slow you down no matter what.
@@samburti8587 If someone told you they were going to overclock their GPU, I think you'd have an idea of what they meant. If another person told you they were going to undervolt their GPU, I think you'd have an idea that they meant something else. You would then use this opportunity to insist that they are the same thing when, clearly, two different ideas were expressed ... and you understood that at the start.
You introduce logical confusion by ignoring the idea that 'performance' can be measured based upon what the person wants to achieve. You would insist that performance can only be defined by fps. This is the tactic of a brute.
Been using similar settings myself (80% power, 900mv at 2625 Mhz, memory +1000). Feels nice to outperform my previous GPU, a 3080 Ti, so much, at lower power draw.
Damn I must have a good card, I can do 2775mhz at +1400 memory on 80% power
@@sorvex9 Do you also undervolt?
@@Goodbutevilgenius Hmm no, I don't see the option to undervolt in MSI afterburner. All I can do is turn down the power limit.
Is that ok for long term while oc the memory ?
@@TheAnugia as long as nothing overheats (which is dependant on fan speed, of course, not listed here), these settings are absolutely safe.
great video. my 3080 was taking about 300-350W so thanx to the sweetspot 80% power I can change to 4090, keeping my corsair 750W psu
Awesome stuff. I have a similar system with a 5800X and 4090 FE. I've set a 70% power limit and haven't toyed with undervolting the card just yet, but I'll go ahead and give it a try. It's amazing how efficient these top end GPUs and CPUs that have recently launched can be, and how it gets overlooked because Intel/AMD/Nvidia decide to push them to the limit.
You definitely need to upgrade that CPU soon lol.
@@ZackSNetwork 5800x is good, not ideal to pair with a 4090 most certainly but its definitely not the worst choice either
70% power limit seems to work great. I have had no issues running games at 4k 120hz, so its nice to know if I ever run into issues, I can just raise the power limits and get a little extra performance.
I'm a newbie and I'm scared to mess with the volts so I just dropped the Power limit to 60-70% and games run amazing anyways with that said do you guys recommend me trying to mess with it its alot of money I'm playing with as a new pc user
Thank you Chef ! I also have a 4090 Msi X Trio, 4°C gain, about 60 Watts and the same performance! THANKS DUDE !
please tell me exact settings that you use
Undervolting your GPU (especially latest brands) should be done by everyone. Great video
actually should be done by the nv before ...
Sounds like the poor her spoken. Lol
No, just don't buy cards you can't afford to keep.
@@kaiakoken wdym
Pretty useful video but the thermal testing was next to pointless with the fans running at different speeds across the 3 modes. Should have locked the fan to a particular speed for all tests.
Another cool addition to a graph would be frame capped at 120. Just to show how much power it draws in usable situations. And maybe compare it to last gen. I guess I could do it… have a 3080 to compare with. I bet it pulls under 200w in some games. Nuts
I found that most games only pull 350w tops. So same as my 3080 10g. I thought about changing my hx750i psu for exactly 0 seconds… I knew it could be close while OCed, but also assumed it would be very efficient at 350 on Tsmc 4n. I was right. Total sys power draw is 350w in several games I play frame capped at 120 on Lg c1.
If you wanna get rid of coil whine, undervolting can really help with you that as well. I know undervolting my 3080 helped a lot with the coil whine.
How do you hear coil whine over your headset and machine guns/explosions/swords clanging + game music? I never understood the complaining about fan noise etc, it must be as loud as a hair dryer for you to hear that over your headset.
@@Cruor34 I hope you realize computers arent just used for games
I don't know how I have avoided coil whine my entire life. I havae gamed with 2080-3080-3090(x3 - yes 3 different cards)-4090 and I have never heard coil whine.
I did some test with my 4090 in furmark.
950mv undervolt gave me 744 fps / 355w / 65c (2800 mhz@ 0.95v +1500 ram 80 power limit)
stock gave me 786 fps / 450w / 73C
full blown OC gave me 796 FPS / 490w / 77C (+200 core +1500 ram 133% power limit.)
so we are looking at a decrease of 6% performance for 30% less power when comparing the overclock to the undervolt.
If we look at it the other way, you would need to increase undervolted card's power draw by 40% to gain 7% performance.
im going to stick to my undervolt, the fans are loud as hell when it gets hot XD.
i also tried just playing with the power slider. at 80% power limit the results were similar to the .950 undervolt but voltage was always 10-15w higher and performance the exact same.
If you undervolt the card, does the fans are loud when it gets hot, or you are referring that you undervolted to minimize noise of the fans?
I had a hard time telling WTF was going on until I realized you were moving the configs around. You should keep the places constant for something like this.
Good video buddy. U will grow
so if i decide to undervolt that means i can keep my 750w psu my current gpu which is a 3080 is using about 330-345w
Kinda wish you would have been a little more detailed on how you do this, the very short clip shows you dragging one slider down to a random number 172, whats the significance of that? Then you drag the whole curve higher which isnt very clear how this is underclocking the card. I want to do this but am almost more confused after watching this video.
Gamers, when the GPU doesn't use the full power potential getting slightly less fps improving power draw: WE NEED TO CHANGE THAT!
Gamers now, when the GPU is using its max potential from the start: WE NEED TO CHANGE THAT!
honestly can get away with 750W or 850W power supply with undervolting, at 950mV, it draws no more than 320W
I have a SFF 850 watt coolermaster driving a 13900k power limited to 180 watts and 4090 at stock with no power limit (450 watts)
System runs great.
A quality 850 watt PSU is fine even at stock, as long as you don't let the CPU go bananas
I'm looking to still run all my games in max setting (1440p), but getting lower temperature. I just try undervolting at .950 @ 2700 mhz with +1000 memory clock and power limit 80% and I have pretty much the same performance than default, but -8 degree celcius with the Benchmark of Rdr2 (yes I know, I love to use in-game benchmark). I think the best, is to keep high frequency, but at the lowest voltage possible. I think .950 is really decent. During the Benchmark, the voltage was .960... I don't know why it goes up a little but, but it's still lower than the default (1.050 I think). I see a lot of people talking about just lower the power limit... I tried and it didn't work. Everything by default with the power limit at 80%, the voltage was like 90% of the time at 1.050 and just get to 1 a little bit. I had average FPS -1, so pretty much the same, but with 1 degree hotter (not a average, I just look the maximum temperature I got during the test).
My fe 4090 CANNOT handle 1500 on memory shame really
@6:20 epic voice crack
lol nice catch.
I do -300 mhz on Core, then find .970v in curve optimizer and pull that up to 2820mhz and hit checkmark.
Running stable for over 10 months on RTX 4090 Gigabyte Gaming OC. 2805 mhz is another great value.
I wouldn't touch memory, memory overclocking on GPUs really makes cards unstable and degrades the chip too quick.
What is "hitting checkmark"?
Undervolting + power limit doesn't make sense in comparison to stock for power consumption. It would be better to run both on 90% limit, or without any.
so what i am getting with my OC is hdr 4k settings in max no dlss either is not over heating but a type of lag. but my point is im getting over 120fps when i first play the game then random drops to 45fps mostly when i inventory and camp site or a sudden change of room. so cpu goes high too but gpu drops to 34%-56 % or worse. just awful. so the card easily handles any settings for awhile but then doesnt handle it. i have a 13900ks cpu too so its not going to stop it.
Great video. Great tutorial! Subbed!
How did you unlock core voltage on AB for the 4090? Mine is locked!
Almost 100w savings but only less than 5% loss in fps
Yup. If I'm not mistaken, the 2nd undervolting configuration has an average loss of 2.21% FPS while decreasing power consumption by nearly 20%. That's like 30 something Watts per FPS lost.
I waterblocked my FE 4090. I just set GPU to use 600wats and overclocked with after burner till stable, uses 450watts with no fans! when running 4k/8k ray tracing. BUT - Then I limit frame rates around 200 so games like Destiny and Diablo barely tax the card/ hence I get good power to performance because it never redlines, not sure the logic works out but seems to offset cons while ensuring I can hit peak OC performance. I would not suggest doing what this guy suggests..., as I can get this card to 35fps on a bunch of games I play 4k/ray tracking and with frame gen and above tweaks I get 70-90 vs. sub 60 stock..... so barely good enough. Jk
Those frametime times are pretty nice
When my 4090 comes in I'm doing this for sure.
Can you please explain why you would drop the entire curve down first then spike up a single point where you flatten, it makes no sense to me why not just leave the curve as close to default as possible for stability sake?
Great video, thanks !
the problem of temps is that msi trio dosent have a vapor chamber
that's why I got the msi supreme
RTX 4090 AMP Extreme, 2.7 GHz, +1500mem, 950 mV, 90% power and Cyberpunk crashes, I believe it's not enough power. Any thoughts?
Edit: After a few tests I just set it at 975 mV and it works perfect.
What's the point in buying an rtx 4090 then in the first place ?
Great video btw! I'm trying to set the curve as you showed in the video, but when pulling the dot above the 950mv only that point goes up. It doesn't level everything out like yours did in the video. Were you holding another key when moving the point by any chance?
That's the way it is have to be. You then hit the Checkmark on main MSI Afterburner window, and it will even the rest of spots.
Make sure to do -X value before doing that (on core)
I do -300 mhz on Core, then find .970v in curve optimizer and pull that up to 2820mhz and hit checkmark.
Running stable for over 10 months on RTX 4090 Gigabyte Gaming OC. 2805 mhz is another great value. -300 is very aggressive though -200/-250 is a safe bet
I wouldn't touch memory, memory overclocking on GPUs really makes cards unstable and degrades the chip too quick.
Overclocking should be a choice, not nVidia default. Most people won't do it, and many of them because they want efficient hardware. The 4090 out of the box is anything BUT efficient. It is running way beyond the efficiency curve into power hog territory. Driver tweaking does not solve this. It does not run like that in linux, and it does not run like that if you are running from an USB stick, it does not run like that if you experience a crash, and it does not run like that if you put it into a different PC. So nVidia made overclocking the stock by pushing 150 more watts into the card without us asking for it. If there was a switch on the GPU that limited it to 300W that would be fine for me, but there isn't.
You sir, have no clue what you are talking about. The 4090 runs cooler, and draws less power than my previous 3080. I can tell, you do not own a 4090. You'd never make such a ridiculous comment. The 4090 is MORE EFFICIENT, than previous Gen. You are simply parroting all the early reports and guesses made BEFORE people actually got the card.
Spongy is correct. The 4090 draws less power than the 3090 and can be undervolted by 20% for single digit performance loss. For which I thank all the voltage outrage over the previous months for achieving. I'm betting they redesigned the card improving efficiency after the backlash resulting in a far better product.
@@mr.spongylikeaboss4987 So glad to hear from folks who actually have the card, since I just ordered one and I'm also upgrading from a 3080.
@@geronimo5537 sorry guys but you are being delusional. The 3090 has a TBP of 350W, and won't draw what the 4090 can. You are perhaps confusing with the 3090Ti. But the 4090 is not the top model of the 40 series so you should not compare it to the 3090Ti but the 3090. But even if it was it does not matter. It is running completely off the charts in inefficiency. By limiting the card to 300W you lose ~10% performance. So barely over 10% performance for 50% more power. That's a completely overclocked card. It does not matter that it has a 3Kg cooler that keep it cool and quiet, it still is not efficient. I'm not asking for much. I'm simply asking for a toggle on the card that reverts this overclock back into an acceptable efficiency range.
@@marsovac well if you want some details. Seems the 4090 will benchmark the same as a 3090 if power is cut down to 40-50% usage. That is the same performance as a 3090 for near half the watts a 3090 needs. The 4090 also consumes less power in general compared to the previous 3090 as well. Which at this point is very well documented. Any comparison video of the two cards is going to show that information.
Great job. Keep it up.
I hate when you find a beast oc but it still crashes the odd time
60% power limit to 70%, is the sweet spot, 80% does basically nothing. 230-350 watts power usage for 60-70% PL, with little fps loss.
um i guess my 4090 xtreme airo that came from zotac is messed up cause it's BASE is the overclock at 2580 and it boosts out of the box without me touching it or the bios saying "we're boosting you" to 2980. I dont understand how this card shows in everything with the base clock as 2580 and boost 2780 then BOOSTS to 2980 (I have logs) without me changing a darn thing in afterburner or firestorm.
This is so helpful.
I really don't get why no reviewers dose this
I'm very curious, is it safe for the long term if we oc memory?
Have you witnessed the same behavior as Ali did from Optimum Tech? His 4090 lost performance with his UV even when the core clock was higher than stock. He blamed the decrease in "gaming clock". I'd be interested, because the way I UV is to get more performance at lower power draw and decreased thermals.
What's a good way to lower temps on this card? I have a msi liquid x and it can get insanely hot.
My entire room feels like it was back in the xbox 360 era with it being a heater.
Great video and clear instructions, curious if you're going to show the smaller form factor build you mentioned in another vid or would that be a while away? Interested to see how the card fairs in those situations.
I know this is an older vdo. Maybe things have changed. I currently have my 4090 overclocked +235 core , +1317 mem , max power , the most ive seen it draw is 560watts. I'm still on my 800psu. I do agree with you that core increases barely gave me gains, I think I saw 2 fps more in Red Dead Redemption on a 120fps average scale. The memory overclock on the other hand gave me like +10 fps.
I'm not sure why I'm drawing so much more wattage than you but I did see one of your benchmarks that showed the Overclock being slower than stock. This makes me question how stable your overclock is, you might actually do better with less overclock and it'll pull less wattage automatically, this would skew your results to show significant gains for the power usage.
Overall though, I'm seeing about 10% gains in overclock. This amount is significant if you are looking at it versus a 3090ti or something it gained 50% performance on. In a resolution lower than 4k, 10% is a decent amount. I guess the bigger question is, other than the concern over powerbills , whats real legitimite use of undervolting.
My answer would be, to be able to use the card in a PSU limited scenario. Not having to spend more money on a new PSU is worth more than the questionable utility bills savings I believe.
It's small energy expenditure to be honest. I 'm old school where our lightbulbs used alot of watts. Turning off lights and appliances can make up for the power savings. Anyways rant over, I'm sure everyone has their own idea over what they want out of their card.
I didn't spend mucho money to gimp its performance. I didn't get an AIB with a huge cooler problem in order to try to reduce the heat by reducing performance.
The perfect spot place I would love an undervolt is if I can get the same overclock. Maybe its overshooting the required voltage at the moment.
Nice video. Why are you both using power limit and undervolt? I strongly advise against purely using power limit, as it uses more power with less demanding games versus an undervolt. But what is the benefit of using undervolt + power limit? With the undervolt it will hardly ever hit the power limit.
Some more information for readers:
I'm running my rtx 4090 at 900 mv with 100% power target. it is a 3% difference, and with my VRR screen, i certainly can not see the difference between 100 fps and 103 fps. With a lot of games maxed out, my RTX 4090 stays below 300 watts. I found that only with Metro exodus, the power usage can go up to 380 watts! With Forza Horizon 5 maxed out, with RT it only uses about 260 watts. This is with the 900mV undervolt.
would have liked an efficiency chart
If I want to “prevent” a fire with my 4090 Trio connector, how much can I undervolt? I want to wait till Cablemod releases the 90° adapter. Thanks!!!
Unfortunately the connector melting issue has little to do with power draw but more so to do with misalignment of the pins due to the adapter being bent. I'm hoping Nvidia will sent out right angle adapters to people for free because that is just unacceptable.
The real problem is the internal foil pins cracking and arcing inside the connector. Basically any horizonal stress can cause this. So its best to keep the adaptor very stress free and limited to vertical bends.
@@DannyzReviews the issue mostly applies to people pulling it left or right. Up and down is less of a problem
Hey do you have a video on PBO and CO overclocking the 5800X? I have one to. I usually just do an all core of 4600 to 4700 but would like to try PBO+CO
You can find that video here. It really isn't an in-depth tutorial video on how to use PBO2 and curve optimizer but I do briefly cover the methodology so that anyone can recreate the same settings themselves.
ua-cam.com/video/wp9OmI7Xbnc/v-deo.html
Great video
hmm, interesting... So the 3 settings to change in MSI afterburner are Power Limit, Core Clock and Memory clock right? You don't have to change anything for "voltage".
That's correct. You don't have have to touch the voltage slider as that's already being taken care of in the curve option.
Or just lower the maximum temperature. MSI Afterburner will then lower the other values accordingly.
Did you notice any reduction in coil whine when undervolting?
therefor a custom fancuuuuuuurve - pause - unpause- wasnt necessary
05:09 fart?🤣
Its a secret movie reference lol
Does VRM quality etc matter when undervolting?
4080 to? My self clock up to 3,1 ghz msi suprim x white very good cooling and case, 125 powerlimit and 88c monster! Hade asus tuf pc 4080, only 2,8ghz and coilwhine
I don't know if it matters, but did you have different settings for the Benchmark? Or was it somehow shots from a different time of day? The shadows look very different.
would that affect the lifetime of gddr? im asking coz im gonna be doing lots of raytracing in vray
I have a Intel i9 108850 running more or less stock I just set all cores to 5Ghz as a minimum and left the regular higher speed cores as is.
I have 64GB memory and 10 Corsair LS13 Led fans and li strimmer lights. I have a RMx 850watt PSU.
If I undervolt and do what you have done is my 850 watt PSU enough with a 4090?
In cyberpunk 2077 if you set the rt on with ultra does the undervolt crash the card?
I don't know why but my msi rtx 4090 suprim only takes 280w on full load and it runs at 2800mhz don't know why and no fps drop and got 1000w psu so I'm not complaining lol
the suprim x might also work cooler because of its vapor chamber!
Power Limit 85/ Target Temp 88/Core Clock +205/Mem Clock +1500/ Fan Auto. Max I see for watage is 380w(down from 430+w) but the clock will just sit ther at 2800+ (I've seen it sit at 2900+ in some gaming Benchmark Tools). I'm done tinkering yay! MSi 4090+5800x3D. P.S. I think Nvidia is setting us up for a "more efficiency" scam with the 5000 series. My card should not be this Stable with a near max clock and a huge chunk of wattage missing(17 Benchmarks cleared with no crashes including the 3D Mark suite).
How does, when undervolted, the power consumption develop when you instead of introducing a powerlimit use a fps limit, like 100 or 90 fps on a monitor with g-sync? The reason I am asking is because I am puzzled which one is more efficient maybe a mix of both fares best? My target is 90 fps, but I am not sure how the results are especially because atm I only use a fps limit, I lack the equipments to test that.
I did the test with borderlands 3, on 1440p, ultra settings.
Stock:
Peak power consumption: 435W.
144fps limiter (via nvidia driver):
Peak power consumption: 354W
144fps limit + UV (2550Mhz @0.900mV): 201W.
how would a bios switch affect this? If undervolting , should it be set to quiet or performance?
Performance is the default, do leave it to default
What if try to use PL like 70% + OC Core/Memory using sliders?
What was your stock Speed way score? out of the box? im getting around 9630 which i think is on the low end for the same card as you. it should be closer to 10k. Can you tell me what you were getting?
I got 10080ish out of the box. Speedway is leaning towards memory OC tho.
What’s ur cpu? And don’t worry about low score out of the box, u can tune it up urself
hello what is your current daily settings? also do you have any coil whine/buzzing?
For my daily settings I will be running with the 2nd undervolted config, where I targeted 2.62GHz at 900mV, overclocked mem +1500, and limited power to 80%. I'm okay with losing a tiny smidgen of performance to save like 20-30% power and lower noise.
@@katates My card's coil whine is similar to yours, and I've seen other samples from others 4090s and it also sounds similar. I think its just going to be normal with these cards, considering how large the die is, how many components are on the PCB, etc. Undervolting does actually lower coil whine a bit because there's not as much power being shoved through those inductors and coils.
pulling " only 285W" ... phew... we went a long way since the "power hungry" Pentium 1st Gen.
I have the same exact card. With the newest drivers the 2nd uv crashes on timespy extreme.
Well I retired it today and it works now. I dunno
@@blancs3030 3dmark does do some weird shit from time to time lol
Hello i need help, when i preset 900mv 2662 mhz 1500 memory clock and i try furmark my pc crush, and my image gets black point, any help? Sorry for my english
Does Msi Afterburner have an issue displaying 1 percent lows with a 4090... i cant get my to display in the games
You have to go to options, then benchmark and set a key like z, a, w, o or another letter on your keyboard then you have the lows
Your curve is messed up on the 80% Power limit
is there no problem in long term when i overlock the memory clock to 1500?
nope
No, they don't have to. I don't do overclocking to have less FPS and worse results in benchmarks by undervolting. If you can't afford electricity, don't buy an RTX 4090. This subject of undervolting is boring.
My both these configs crash my games lol
Does the brand even matter when undervolting?? I plan to get the 4090 Suprix
Not really tbh. Solely depends on silicon lottery
that MSI 4090 is thick!! is it 4 slots?
3.5, so ya basically a 4 slot card
Rtx 4090 is always slower on a undervolt this isn’t the rtx 3080
Would this help with the melting problem?
You just need to plug it in snug and it will not melt.
@@JagsP95 💀
@@R420ISH Not even kidding man, watch gamer nexus video and its insane that it was user error.
How are you getting these high temps I get 55 on my 4090 while gaming
What resolution are you using? He is using 4k and it's pushing the card to work more intensive
3440x1440@@eduardbotnari
Dont need to do anything most people don't care leave default.
would the results be around the same at 1440p?
Less power draw and temps.
I’ll just keep my 4090 Fe on stock.
If you can afford a 4090 you don't need to worry about electric bill lol, except if you live in san diego lol
Very sceptical about the no performance loss from the thumbnail. I’ll stick to the OC
Why are you skeptical? The video provides benchmark results which demonstrate a 1 FPS AVG and 1 FPS 1% low difference between stock and 80% power undervolt. And if you want, you can just test it yourself.
@@Goodbutevilgenius for staters why put +1500 mem on the under volt cards a not and on the stock is he’s trying to compare like for like. This generation it’s the memory that has big impact on performance compare to the core. So the results have been skewed to show little to no loss. Not saying the difference will be big but there will be some performance drop off. So by saying no performance loss in the title thumbnail is false.
I do this on my 3080ti, running at 80%, there us a small perf loss but I save upto 100 watts and it runs cooler
I buy rtx 4090 tuf non oc then i flach Strix Bios on it
This is all too confusing for me.
I like how he moves the graphs around just to mess with us. 100% deserving of a dislike.
bruh jus cap ur fps it ezzzzzz just a waste of pwr 2 go over ur monitor refresh esp if its only 60 u will get lik 200v draw that way
Buy 3050.
I pass 420watts with my 4080
Too bad the 4090 is pretty underwhelming vs cost... lol
dlss 3 and frame gen more than make up for it
how is sub 70c temps pushing the gpu to its limit?
You have the least annoying intro of just about anybody.
I’m not throwing away performance. Setting a custom curve is only really necessary if you’re power limited or having stability issues.
At the end of the day its up to you. I personally just felt that If I can save on power while getting the same exact experience as stock, I'll definitely make the trade. Plus I also game at a capped framerate on my LG C9 to stay within the gsync threshold.
@@DannyzReviews I’m old an lazy. I’ll mess with the curve if it becomes necessary. I used to have to do it twice almost every time I turned on my system with my 2080 Ti’s because the voltage curve was buggy with those and wouldn’t ever save correctly if I restarted. They were severely power limited and unstable with the auto curve, so it was pretty much necessary. The upside is I got really really good at tuning a custom curve 😆
Too much useless talk.
Bad results. My 3080 goes from 320 watts to 250 with 97-98% of the performance and -10 degrees on the core.
If you look at the average graphs, there's an overall reduction of about 20% power consumption for a 2.21% in FPS loss when compared to stock.