The Brewing Problem with GPU Power Design | Transients
Vložit
- čas přidán 26. 07. 2024
- Support our in-depth testing, journalism, and analysis! Grab a Modmat, mouse mat, shirt, or more at store.gamersnexus.net/ (through 6/25, 10% of ALL store revenue ALSO goes to a charity handling technology education for kids, e-waste processing, and refurbishing).
Grab a GN Tear-Down Toolkit with 7-year warranty! store.gamersnexus.net/product...
Or a GN Hourglass "Silicon Waste of Sand" T-Shirt in cotton - store.gamersnexus.net/product... (Tri-Blend also available)
Black & Red 'HUD' Desk-Sized Mouse Mats in stock & shipping! store.gamersnexus.net/product...
Power transients from high-end NVIDIA and AMD graphics cards can cause PC shutdowns. We explain what they are, measure them in the 30 series, and discuss their impact on the upcoming NVIDIA RTX 40 series GPUs (e.g. the rumored RTX 4080 and RTX 4090).
NVIDIA’s RTX 3090 Ti consumes more power than any consumer GPU we’ve ever seen. The power draw of this card prompted investigation into how much power 3090s, 3080s, and even the venerable 1080 Ti consume. Transient power spikes from these cards can ask for 2x and more than the normal max power target. PSUs like the Seasonic PRIME line and the EVGA GA line might have trouble with these transients. We explain what these transient spikes are and how they impact PC building, and particularly, how much you need to plan ahead for future GPU purchases if this problem doesn’t get addressed by NVIDIA and AMD (or the PSU manufacturers) adequately.
RELATED VIDEOS
Watch our PSU Reviews, Deep-Dives, & Explainers Playlist! • Good vs. Bad Power Sup...
RELATED PRODUCTS [Affiliate Links]
NVIDIA RTX 3080 GPUs on Amazon: geni.us/tMuD
If you want a PSU that handled Transient spikes in our testing, check out:
ASUS ROG Strix 850W Gold PSU on Amazon: geni.us/u7AsXLD
Gigabyte GP-AP1200PM 1200W PSU on Amazon: geni.us/i9FFjYi
Like our content? Please consider becoming our Patron to support us: / gamersnexus
TIMESTAMPS
00:00 - GPU Power Requirements
01:30 - A Concerning Future
04:00 - The Issue
04:48 - The Problem with Average Power Consumption
06:15 - Overhauling Test Processes & SFF PSU Risk
06:47 - NVIDIA or PSU Makers To Blame?
07:52 - What Transients Actually Are
09:39 - Transient Load vs. Response
10:12 - GPU Power Draw is Changing
11:20 - How Power Overhead Works
14:14 - Over 40 Combinations of Hardware & Software
16:05 - Test Data
18:30 - Investigating the Power Supplies
19:17 - Why Small Form Factor is a Problem
20:05 - GN's Power Validation Setup
23:44 - Conclusion & Reality
Check
** Please like, comment, and subscribe for more! **
Links to Amazon and Newegg are typically monetized on our channel (affiliate links) and may return a commission of sales to us from the retailer. This is unrelated to the product manufacturer. Any advertisements or sponsorships are disclosed within the video ("this video is brought to you by") and above the fold in the description. We do not ever produce paid content or "sponsored content" (meaning that the content is our idea and is not funded externally aside from whatever ad placement is in the beginning) and we do not ever charge manufacturers for coverage.
Follow us in these locations for more gaming and hardware updates:
t: / gamersnexus
f: / gamersnexus
w: www.gamersnexus.net/
Host, Copy Editor: Steve Burke
Writing, Testing: Patrick Stone
Video & Production: Andrew Coleman
Additional Video Work: Keegan Gallick - Hry
Support our in-depth testing, journalism, and analysis! Grab a Modmat, mouse mat, shirt, or more at store.gamersnexus.net/
Watch our PSU Reviews, Deep-Dives, & Explainers Playlist! czcams.com/video/yDX_1PWUWdw/video.html
Love how the intro of the video is edited like a Dateline NBC report.
THANK you for this video! We've had customers GPU's returned to us because they exhibit that exact behavior, tested (Corsair AX 1600's), found to be working flawlessly sent back and forth a couple more times before we've figured out that their PSU is (although above "spec") not handling the loads some of these monsters pull under spikes.. GPU's should have transient spike recommendations included. Easy as that. Bad marketing for Nvidia/AMD? Ok, we (small retailer) do not give a f*ck, current system is misleading. AND! A lot of customers don't enjoy getting "up sold" on PSU's.
Great video Steve and his team!!! Glad you guys made this video!
I am a Mechanic (40+yrs), and I have a general rule. Your battery should be 2 CCA (cold cranking amps) per cubic inch of engine displacement. I follow the same rule when selecting power supplies. I make sure the PSU can handle near double what the system can draw. Your PSU will be most efficient at around 50%, it will also run cooler and quieter, and it will last a lot longer.
quick correction: the magnetic field is what creates the charge, not the other way around.
Edit: I would like to see what effect undervolting has on these spikes if any at all. I know you can get some interesting behavior out of a GPU by undervolting a little too much.. but doing it just right seems to make it more stable since clocks are not constantly shifting.
Electrical Engineer here, I have to say, what a superb animation of how current clamps function. This piece of journalism is outstanding, you guys rock.
Thank you! We'll let Andrew know you liked it!
scrolled down to say the same thing, I've ALWAYS wondered just how those clamps work their magic! sick job GN crew
@@GamersNexus Industrial automation electrician here, and I wanted to say the same thing! You might be inspiring the next generation of industrial controls tradesmen and women! lol!
I'm a computer engineering student, and I wish I had seen that current clamp animation for my lab courses - it really is fantastic!
@@GamersNexus Andrew really killed it on that animation, his animations really add a whole new level of explanation to the videos over the last year or so.
The level of quality content your team is putting together is down right impressive. You guys keep this space in check and I truly respect what you all are doing.
Thank you!
@@GamersNexus no thank you
And completely relevant. A lot of trust in this community
Honestly.. what a great fucking video- truly.
So well said...i absolutely love these types of videos from GN. In-depth and very news / informative type for consumers...
"We adjusted for the latency in the hall effect sensors........". You guys are truly an amazing group.
@Lurch7861 Lurch7861, My sincere apology for making a comment that upset you so much that you felt compelled to call me names.
@Lurch7861 ok buddy you sure are very smart
@Lurch7861 No, measurement latency is not a college physics thing. Heck, in my CPE labs, which were the same ones the EEs took, we never bothered to correct for it. Quite frankly unless you're looking for something extremely specific, it's not worth the pain.
Wall power is atrociously noisy as is, and on the DC side we aren't doing weeks of data analysis. We're fixing the problem! Heck, most of the O-scopes I used were CRT. There was no data export. Even the digital scopes I think the most I ever did was export screenshots. And my senior design project was on power monitoring...
@Lurch7861 : It's not a matter of having the education to know that you need to do something, it's about having the time and money to spend on doing it.
@Lurch7861 My background was hard science and I had to take more than a few college physics classes. Never was taught this…
This is an electrical engineering concept.
From an electronic point of view it makes much more sense to solve the problem on the GPU side where the load characteristics are known and you can even actually actively control the spikes.
This can happen at any device inside PC, so its better to be solved at power supply.
@@TheSkace The correct solution is that GPU manufacturers stop lying. If GPU manufacturer says "290 W" but oscilloscope shows 620 W, then GPU manufacturer is lying about power usage.
When all manufacturers tell the real power usage, getting the correct PSU is easy.
With the current lying trend, you cannot know tomorrow for sure if you need to get a PSU with 2x, 3x, 4x or even more power compared to the number that GPU manufacturer prints on the box.
If GPU manufacturer doesn't want to print 620 W in their spec sheet, they need to add enough capacitors to get the power need closer to average.
The gpu power delivery system is relying so much on the PSU nowadays it means the GPU CAN BE MADE CHEAPER because less components
@@urbansnipe That's absolutely true that building a GPU that has power spikes is cheaper. However, when GPU manufacturers lie about the power requirements of those cheaper GPUs, selecting a PSU is turned into gambling. When a single manufacturer says "290 W" do you really need 750 W, 900 W or 1200 W for the GPU? How much could any specific manufacturer lie?
I'm absolutely fine with whatever spikes any GPU has as long as the manufacturer *honestly* declares the needed power levels instead of average power usage which doesn't include actual power required for the spikes.
@@MikkoRantalainen true i feel the same way
Now PSU manufacturers or brands can capitalise on this issue by advertising how massive their Caps are in their unit.
Nominal Load : 850W
Transient Load : 1700W@20us
At a bare minimum, ATX 3.0 + PCIe 5.0 certification of PSUs (with independent verification) will go a long way to educate consumers here. SFX PSUs may perhaps give way to SFX-L..
@@ikjadoon Nah, there's too many people willing to spend hundreds more on a part to save a cm of space. I'm sure Corsair will find a way to make their SF series well over 1000W, if it means they can charge a boatload for them compared to normal ATX psus
Im hoping they also start to put more capacitance on the GPU board. Loading up the caps in the PSU destabilises the 12V rail which will reduce the stability of the whole system, not just the GPU
The good news is capacitor technology is advancing relatively quickly, so it will at least be possible to keep up
@@toby1248 why would loading up capacitors destabilize anything? Aren't capacitors very slow to charge compared to their discharge speed?
@@LtdJorge all capacitors charge and discharge at the same speed, somewhere between a millisecond and a nanosecond for a full charge. But the voltage drops linearly as they discharge, so the more you load them up the more the 12 volt rail will drop below 12V.
I find it kind of interesting that this is just now being discussed for computers. I did car audio installation professionally for over 14 years and this was something we dealt with on a daily basis for huge car stereo installs. In fact it is still something that is dealt with today for multi amp power hungry car audio systems. We had to install large capacity capacitors in the system to "suck up" high transient power demands to keep from potentially hurting a cars alternator. You can visually see what these high capacity caps would do because the headlights wouldn't dim, or not as much, under large bass drops which would draw the most amount of power. Large caps are designed to quickly deliver large amounts of current very quickly while a cars alternator can provide large amounts of current but just over a longer period of time. A computer power supply is roughly the same. It produces a fixed voltage with a variable current output, very similar to a cars alternator, and neither is extremely suited for potentially a massive current draw in relatively short amount of time. For cars it has always been you upgrade to a larger alternator, or more than one if needed, you add several farad caps or both sometimes. It is getting to the point where either manufacturers are going to have to start recommending more accurate power supply requirements to account for this issue or someone is going to have to design a power bank of large caps that go in between the power hungry graphics cards and the power supplies.
Electric amplifiers have had this issue for ages, and the solution is a big capacitor.
I was just going to make this correlation - RMS vs PEAK and what not but you nailed it.
@@StephenGillie The best solution is an elegant one. Graphics cards should implement their power limit properly. It seems like really messy design to me that you can run a load of never above 200W, yet for less than a millisecond it might spike to more than double that. This can be avoided with sane/smart design. But of course the GPU industry is the epitome of superficial power craze, so we cannot expect them to sacrifice any potential in order to avoid having to oversize your power supply.
The solution to transient spikes is not just huge capacitors, at least not on the power supply itself. The power supplies (switching DC/DC converters) have a maximum amount of output capacitance that they can deal with before they go out of stability and start oscillating (which will burn out components). A power supply manufacturer can't know how big the spikes will be for every GPU that you put in your system. That's where the board designers of the GPU come in, they should make sure that they have enough filtering on their power supply input so that they don't trip the OCP. They are the ones that should place the large capacitors on their board. But ideally you don't want to have these spikes at all, and you also place inductors in the line to suppress these huge spikes, so just a lot of caps alone could be detrimental to system stability.
Yup, back in the 90's I had a cap the size of a Fosters can alongside my 1000W Phoenix Gold amp. Ahhh, the good ol' days...
Graphics cards need to have enough capacitance to cover the transient spikes they generate and offer a smooth enough power draw that at no point does the card go beyond the publicly stated max power draw. If they have a device that does not or can not do that then it is out of spec and they need to publicly acknowledge this phenomenon and publicly report the max transient power spikes so pc builders can ensure they pair a power supply that can supply the demand. Anything less is not acceptable. This is clearly a design problem on the graphics card manufacturer's side. How many consumers have dead power supplies only because graphics manufacturers incorrectly designed a card and did not report the true max power draw? Repeatedly causing customer power supplies to repeatedly trip OCP and the customer not knowing what it could possibly be given the power budget allotted to the components and the power supply capability.
This is exactly it!
I would think so too.
This was my thinking since these "mystery black-outs" started and still is. The power supply might be the better place for a huge capacitor, though. But it seem like it doesn't actually need to be that huge.
i came here to say something like this and this is perfectly summed up. it is 100% upto the card manufacturers to disclose the max power draw even if it is "transient" because its still the card that is causing the problem. its a massive copout for such a huge company to blame another company "because the psu shuts off" nah my dudes. make your cards within spec or disclose the maximum draw instead of pointing fingers. we have never had problems like this in the past untill nvidia started making these cards that seem to double power usage every generation. as a pc builder it would be super embarrassing to complete a build and have the customer say that there are these kinds of problems. reasons like this is exactly why i prefer radeon cards at the moment. they require much less power for similar performance in most settings.
I think part of this problem is software in that today's GPUs are more aggressive in ramping up and down for performance reasons.
2017: GPUs are causing massive worldwide power draw.
2018: Countries introduce laws against GPU crypto mining.
2019: GPU prices 5x higher than the systems they're built for.
2021: GPUs affected by worldwide component shortage.
2022: GPUs are frying entire systems due to unprecedented / sudden power consumption.
Man, GPUs are like the bad kid in the neighborhood.
But us consumers NEED the overkill for gaming and streaming so we can compare our e-peen.
@@mosando No you don't.
Never had this problem I bought AMD. :p
@@Life_Is_A... have you heard of sarcasm?
@@Life_Is_A... yes we do
This needs to become a part of GPU reviews from now on.
yeah, if the normal power draw of a 3090 ti is as high as it is without even factoring in the transients, then I can't imagine what the 4090 ti will be like.
@@plasmaoctopus1728 Especially if the 4090 Ti is shunt modded 😅😶...
This channel is much more than a “gamers nexus”. Thank you so much for your scientific, transparent and objective media reports. What you’re doing is invaluable.
🎯
Seriously, I honestly feel like his the name has been outgrown by what he is doing now. This is serious pc builder news. Although it all does tie back to gaming in the end I suppose to keep his scope in check.
You have 69 upvotes, and while I'd like to be one of them, I'll take one for the team.
Gay.
I would dare to say GN is a bit like what savagegeese is to car reviews.
So if you like honest people applying no-bullshit non-sellout scrutiny and are even remotely interested in cars, check him out. (He also has livestreams where he mentions how hard it is to do when you are not being the industry's monkey.)
We need another episode, a sequel to this one, focusing on daisy-chained cables and multi-rail vs single rail configurations and how they relate to stability under transient loads
Had a AIO Vega 64 card on a x370 MB and a Rosewill 1350 PSU . Started having this issue occasionally in 2020. Started being a more consistent problem in 2021 where I could reliably shut down my system by loading a stage in Destiny. I was beside myself trying to troubleshoot, crossing my fingers with each AMD driver update. When GPU prices came down from ridiculous to just gut punch levels in 2021 I desperation bought a AIO 6900X card and a B550 board, assuming it was my card struggling to keep up with newer games as I play in 4K. The problem seemed better at first, but I about had a fit of rage when it happened again for the first time. I would play for several more months clenching my chair from the anxiety of knowing my next crash could be at any moment, and I could almost never consistently reproduce the problem. It wasn't until I saw you mention this while you were researching this video that I finally had my "AHAA, this explains everything" moment. I switched PSU to a Seasonic 850, lower power but much higher quality. After the change it's been smooth sailing since. Huge thank you to GN, I had just thought "I definitely have enough power", and had never considered the OCP angle. Will put a lot more effort into choosing power supplies from now on, and you're spot on in how frustrating it is knowing that each of your components, PSU, graphics, and MB can factor in to how this problem manifests. The average consumer could never get enough info to make decisions on parts in regards to this issue and even most independent reviews would struggle to find the time, resources, and motivation to delve into this topic. Great on you for leading the charge, and this level of production is immensely impressive for an independent group. This channel has become a necessity for being the consumers voice of reason to the manufacturers and developers. Unfortunately for me I noticed your work here too late to save my wallet, but now I know better and am a full subscriber and follower.Thanks!
Yes! Perfect comment :)
This is the most well timed video I have ever experienced. I was literally just telling a friend that his crashes might be from these "power spikes," and he was not understanding what I was trying to say. I'm hoping this video with all your graphs and analysis will make more sense to him. Thanks for the breakdown!
What sucks is a 1.2KW or higher PSU would more likely than not solve the problem completely but that is one hell of a cost to pay for system stability that should be a given if the GPU didn't go overboard on the draw randomly. Locking the FPS would also likely make it not happen and in the majority of games that wouldn't even be something that hurts the experience, but in several titles you want the highest framerates possible so you see the most up to date frames at all times to be competitive and in those titles a new PSU seems to be the only option.
Hmm, I do have random system restarts, now that I got a 3060 ti with a 550W PurePower11, but they happened only when not under any load, and changing the PCI-e setting from "auto" to 3.0 seems to have fixed them. Or at least I haven't had any restars since yesterday morning, when I did that. Although they only happened once every 5 if not 10 hours, so can't say for sure yet.
@@bdhale34 This only affects some GPUs and systems I reckon though. Because I'm on an 850w PSU with a card that draws up to 430w constant power and I've never had any sort of shutdown or instability in nearly a year and a half.
I view the issue as sort of like a hypothetical car that locks up when the RPM reaches a certain threshold because it requires more gas than the fuel pump can push out so the fuel pump cuts off for safety.
Even if the car OEM picks a fuel pump that can supply enough gas for the average RPM, it won't be good enough for when the driver has to floor it.
However, just because you pick a fuel pump with the capacity does not mean it can push enough fuel fast enough, so you could have a car that doesn't shut off but does essentially run on fumes when flooring it.
@@zeratulrus142 I had to change that setting to avoid issues where my PC wouldn't have video when starting up.
Please - If you want to up your game on this, go look up the Middlebrook stability criteria and start reviewing power supplies by their impedance and crossover frequencies.
You're great at this. I think you're one of the only reviewers that might care enough to actually go do it, educate your viewers on why it matters, all while keeping it entertaining.
Middlebrook is the first step basic method for determining stability and impedance matching between power supplies (like the main PSU and one on a board).
FYI, I'm a professional in this field (not specifically computers), but I did work on a motherboard power supply for the first Sandy Bridge early in my career.
You covered it a bit, but everyone thinks capacitors are holdup. They're more for transient suppression and impedance. Those tiny caps on a 1V 100A rail aren't going to hold up Jack. The big caps on a power supply rail are better, but they'll only respond between roughly 20 Hz and a few kHz at most. Your small form factor supplies are pushing it a bit higher (50 Hz or so). What's more important is that higher wattage supplies is pushing the base impedance down and the crossover frequency doesn't matter as much; so stability is easier to guarantee.
The 30x Nvidia series die has multiple full to almost zero load steps with durations of a few microseconds within a single frame. You can literally see a semi consistent pattern at the scene's frame rate if you tap between the on-board supply and the GPU. The on-board supply can take care of most of the transients in the kHz+ range, but those frame-harmonic frequencies (approx frame rate to a few multiplies) are smack dab in the middle of the bandwidth between the cards power supply rail and the main system supply and you have control bandwidth and impedance matching issues between the two.
Ironically, power saving efforts for high performance cards in the future will make this worse. You want it to blast out a frame, then sip juice. It's not practical to store enough energy on a board to smooth it out.
Edit: Also, you mentioned something about the phase shift between the current sensor and the voltage. That phase shift may be 'real'. I'd suggest calibrating it out on your probes measuring against a linear load. If you put your scope in 50 ohm impedance mode you can probably cal the delay between voltage and current while measuring the current into the voltage probe itself, just using the cal rail on the scope. If your current sensor resolution is too low you'll need a signal generator to do that.
I really hope that Steve and the team find your comment and read it. GN strikes a great balance between in-depth reviews while keeping it accessible for most people, so I'm sure they would be able to incorporate all this into their videos.
upvote this above comment
If CZcams would have a Reward Comment system, I would reward you. This comment is gold.
Someone get this comment pinned, get Steve and the team on it!
Could you elaborate on the small form factor PSU's possible impact on being able to handle transient spikes in regards to the frequency you mentioned?
Twenty years ago, when I was in high school, car audio had a similar problem drawing down voltage when the bass hit. The "solution" was adding massive 12V capacitors with claimed 1-2 farad capacitance. A lot of them were junk. I wonder if we'll see similar snake oil for sale in the computer market soon.
I was sitting wondering about that. There *were* capacitors you could put in-line for audio cards. Those, well, *worked*, but didn't make a difference because, shocker, the audio cards did it on their own and didn't really have massive amounts of differences in demand anyway.
Now I am starting to wonder if those products might work for this sort of thing. Just buffer it juuuuuuustt enough. Given that it didn't produce negative issues to it when tested for the audio cards.. erm. Maybe it would work?
Seriously a ducttape dirty hack.
I loved that stuff. I was only 12 but my bro and brother in law had some cool setups. My brothers S10 with huge subs was nuts. Everyone should feel the sensation of that much bass in such a small cabin
I'm so happy someone is covering this issue. Every tech channel should be promoting the hell out of this video and help get this problem under control.
Just want to also say that the animated depictions were both really helpful and well done. It’s easy to take those for granted, but there’s probably a lot of effort that goes into those, too!
I came to the comments to say exactly this. At 21:05 they didn't have to go through the methodology and explain how it works, but they did. Not only that, they put an animation together to show how the instruments measure the EMF to come up with their readings. That's not a small deal. That takes a lot of extra effort and I said wow, that's really well done, out loud when that came up.
I noticed and was going to say the same thing. Really high quality animations and a fantastic explanation of how the tool worked. Kudos to the GN team.
If transient power spikes are going to be accounted for in future component reviews, I am INCREDIBLY grateful. You folks rock, GN!
I was aware of power spikes in GPUs, but until watching your video, I had no idea they were this severe. I'm getting ready to overhaul my PC and my failing old Corsair PSU needed a replacement. I've just ordered a PSU with more headroom than initially planned to prepare for newer cards and I'm sure it'll save me a lot of trouble in the future. Thank you GN. You're the only mainstream hardware reviewers that are covering topics like this with actual data. For how small this operation is, it's incredible how far ahead your data and methodology is versus much bigger hardware reviewers.
Best PC detectives on CZcams, maybe even the internet in general. Appreciate you guys, and all the extra work you put into your testing and videos.
It’s great to see more people getting in on the whole transient issue. First encountered it when Igor‘s lab made it the focus of their power draw and supply tests but spikes and transients will only get more important to look at the higher power draw we see and the more companies rely on boost clock algorithms and sub millisecond control algorithms for power limits and current draws
Speaking of issues... there will be fires in some older houses or houses with often sub-par repairs like in hurricane zones. It's pretty messed up when you need an electrician to do a safety check before you plug in your new GPU, will Nvidia give 4090 buyers a voucher for inspection and testing?
@@Mr.Morden as far as any business is concerned the quality of the wiring of your house is a you problem, this includes the company that supplies your electricity
@@Mr.Morden Not to forget the 110V Powergrid in the US isnt even that great... For me its always funny when people on reddit tell me that i NEED a UPS and iam just "Mate iam german we have 240V and unless you live somewhere far outside and you are the last person on the line and its not burried in the earth or you do super important work you usually dont use them here"
@@mrn234 quality of powergrid in the US is largely down to the state that you reside in. I've never had issues with power in any location I've lived in, and I've lived in the middle of nowhere as well as in the middle of a city with a population of 9 million people.
It's the way how these companies "cheat" better performance, just let the GPU spike for milliseconds so the average performance is higher. Spikes so minor but also too severe to some PSUs, if not bought mindfully. But how should a normal person know about it, to avoid the problems some 3090s had? That's why i like GN so much and people like Igor but also Hardware Unboxed.
GN is straight up making video essays with all of their expertise about tech put into it. This is so good. Props to the whole team. Keep holding the tech giants accountable.
This video deserves far more recognition. The production level, the overcommitment to accuracy and honest transparency, and the attention to detail shown here are nothing short of superb. The GN team is doing amazing things for the PC industry as a whole, and manufacturers should take note.
Gamers nexus putting in work! So much respect for doing proper reviews and in depth articles. This is the stuff the industry needs! Love from the UK
i was not expecting the editing to be so good on this video. not to say your regular videos are edited badly, but this is above and beyond with all the animations and stuff. well done, and great information in here
Damn that intro is a different vibe. Good to see discussion on this topic!
the "looming doom documentary" soundtrack really hits hard.
This is amazing! Thank you so much guys, for these videos! You are doing an amazing job! All the team effort, the researching, the information care, data and information management, all treated with care, quality and responsibly. I'm so proud and amazed by the team :') Respect! Blessings for the whole team! :D
This was a fantastic analysis. Really love this level of information. I do wish there were AMD GPUs mixed in with the Nvidia graphs. Having just worked on a 6900XT build I was wondering how it compares. Thanks so much for all your team does.
yea, would be nice to know the general difference between AMD and nVidia in that regard.
Moral of the story: Everyone needs 1200w PSU's, separate and isolated breakers, tesla solar roofs with power banks, and a dedicated nuclear power generating facility to accommodate next level gaming hardware.
You forgot home and contents insurance when the fire takes your home and assets 😉
No, replace the fuel rods with xx90/Titan cards and you can recover something like 80% of the power
Well...I got the 1200w psu so far.
I have a 1600w psu. 😎
@@LawrenceTimme I almost got one, but I hesitated until they sold out or the sale ended because they put a different plug on them. Does yours have a different plug on the cord from the wall?
I had this exact same problem. After waiting for hours in line to get into Microcenter when the rtx3070 launched, the card I brought home wouldn’t work on any of my systems, and kept tripping the power supplies internal wattage limit. I tried for over a week straight to troubleshoot the issue but unfortunately had to return the card I worked so hard to get in the first place. So happy someone is finally looking into this, and that I wasn’t just crazy. I hope this means the responsible parties will take a look at this problem and work to address it, as it is just absolutely infuriating to deal with when it comes up.
Whats your PSU?
@@Whatthevortez ^ this
You gotta drop the 550watt PSU guy ,I find 750 minimum requirement
@@badcommentbot8349 it shouldn't need to be the minimum, the 2080 ti (for example) could run in systems with 650w and it was perfectly fine
@@chitorunya yeah but this isn't 2019 anymore.
Yes my dude. I have most definitely learned a lot from you and your amazing team of professionals. I really hope to continue learning and understanding with you all but short of taking classes this channel is a god sent.
Thank you all!
I can't tell you how much i appreciate the countdown bars on the side of your graphs. I feel much less panicked to read them when i know exactly how long i have.
This honestly explains so much about my early experience with my 3090. I saw on nvidias website they recommended 750W for a 3090 with a 10900k so I decided to get a 850W PSU to give myself a little headroom. It didn't matter at all as I was constantly black screening cutting out during gaming which I later learned was OCP triggering. It was super frustrating as 1/3 of the people I talked to said I didn't have enough watts, 1/3 said I was borderline and 1/3 said I had more than enough. Long story short is I had to move up to 1000W to stabilize the build and now I believe the reason why is because of the transient spikes you guys are talking about. I really appreciate your video highlighting these issues as I thought I was crazy for a long time in regards to this.
I'm rocking a Corsair RM850x for a 3090 FE and 12900K non-overclocked. I'm scared to even touch clock settings with these kinds of spikes. Probably going to have to upgrade the PSU at some point. My first gaming setup in 2012 was on a Corsair 600 watt PSU. It's crazy to think how ridiculous things have gotten the past several years.
@@braf7349 my wife's RM850x finally started showing it's age after about 4 years with a 2080Ti. It started black screening in games that were only moderately demanding. Monitoring rail voltages in HWInfo, I could see random 0.3-0.5v drops on the 12v rail, even without any real load. I didn't feel like having to deal with the whole warranty process, so she inherited my old EVGA G2 850w (5950X/3090 setup) and it's still going strong for her.
Good luck with your setup! At least a PSU is a pretty easy upgrade if you do have to swap it out.
The 10900k chugs around 200-250W on turbo, so a 350W 3090 with spikes at 700W might indeed trip a touchy psu. (Especially if you have a simultaneous spike on the cpu)
Replaced my 1080 ti with a 3080 ti, had a 750w psu which should have been enough but it wasn't, system ran stable stock when I went with a 1000w psu, likely have some headroom now but I prefer no crashes vs a few more fps
Poor quality or defective PSU? Also undervolting the card saves so many watts.
I am an electrical engineer, sometimes dealing with transient issues in an industrial environment where we work in kW and MW. I am also a keen audio enthusiast and know for example that a power amplifier is very dependent on its power supply's ability to deliver power at higher frequencies. I thus understand your review and kudos for pointing these issues out to prospective buyers. What the industry needs is a rating system for PC power supplies, in their ability to handle short term overloads. Premium manufacturers like Corsair can then use this rating system to distinguish their power supplies from the vanilla variety.
The issue with a rating system such as the 80+ certification that already exist for PSUs is that it becomes vague and eventually turns into a marketing label overtime. Unless there is a federal standards I think we will just be seeing more of the same.
I’m more on the side of putting bulk caps directly on the GPUs to help deal with the load personally.
@@joshuabenitez3260 I think ATX 3 is intended to address this.
Isn't it what these multiple agencies like the Underwriters Laboratories (UL) in the U.S. are here for to abide by their ratings?
I would side with @thelol1759 and suggest that the problem needs to be solved on the GPU side. The PSU needs to have solid bandwidth but there's usually a foot plus of cable between the PSU and the GPU. In the world of sudden, extreme current loads, the rail could be collapsing on the GPU side before the PSU has any chance of responding.
CPUs have had spike current demands in excess of 100A for a long time but their VRM circuits are spec'd to handle those transients and not pass them on to the PSU so severely.
GPU VRM circuits, and particularly their bulk capacitance is just not up to snuff if they can bring down a 12V rail that's basically all theirs.
You have put so much work! Thank you and great job :)
I went thru 4 RTX 3060ti’s before I realized this issue. I had a 600 watt power supply which is enough if you go by the manufacture standards. I now have a 850 watt and have no issues. So I think it was the transient spikes that was frying the cards
The upcoming ATX 3.0 specification, which was finalized in April, is specifically supposed to address this issue. So, I'm really surprised there was no mention of it here, at all. I've been eagerly waiting for some kind of news from Gamers Nexus, about when ATX 3.0 power supplies might be expected to start rolling out, or if the RTX 4000 series cards will have the new 12+4 GPU power connector on them, which is part of the ATX 3.0 spec. Please ask one of the power supply manufacturers, about an ETA on ATX 3.0 PSUs. I'm very eager to find out, since I'm planning a Ryzen 7000 system.
They'll surely have the 12+4 connectors on the FE/reference cards, but are unlikely to require it of the AIBs.
The big question is what the PSU requirements will be. I'd assume they'd say something like '4090 requires a 1000W PSU, or an 800W ATX 3.0 PSU'. Also, as you noted, its unclear if the ATX 3.0 PSUs will be ready for an October 40 series launch. It would be really dumb to buy at launch if you have to upgrade to an ATX 2.x PSU that's becoming obsolete.
Looks like you are expecting people go out to change their PSU with ATX 3.0 spec. Tell it to those people with no knowledge, or those with 1000W+ Platinum rating.
I doubt those with some years old 750W Gold PSU or above will do that so this info is very useful for the years to come.
This guy asking the smart questions. Where's atx 3
@@JorgeAlarconL its very relevant info... Whether you are too ignorant or cheap to buy into the new standard is not relevant.
@@Xirpzy Assuming that everyone should buy into the new standard and that anyone can afford it is actually ignorant. A good PSU stays a good PSU for as long as there are devices it can run, and people either don't see a reason to change their PSU because of a new standard or they don't want to change their expensive high end PSU just because of a new standard. Some people may not even be able to afford a decent PSU at all. Your comment sounds very pretentious as if you're saying "lol just buy it you poor pleb". Very very disrespectful.
This was very well thought out and tested. More in depth videos like this would be great! This tackled something I have experienced and often oversized PSU to handle the issue. Now I can understand what is going on. Thanks
So blown away by the through nature of the research that goes into these videos! Content like this helps us all, because it informs the consumer and puts pressure on the manufacturer to improve their process.
Possibly we need some sort of regulation for power spikes like this , similar to EMC regulations ? A PSU should be expected to be able to handle spikes yes ( due to things like inrush) but not to this level , not in the consumer space.
Most consumer PSU's are generally built to the ATX specifications for power supplies. That spec doesn't really address any specific requirements for transients. However, it does say that DC output voltage must remain within a certain range under ANY load conditions that are within the specified output ratings of the supply. So, if your supply says it can do 40A on a +12v rail, the ATX spec doesn't care if that is a nice steady 40A or if it switches between 40A and 1A every microsecond. It expects your regulation to be within spec no matter how crazy the load cycle is. This is unrealistic, but the spec is silent on transients. Transient requirements for the rails could be added to the ATX spec, but failing that, GPU and PSU manufacturers would have to publish their own tolerances for transients and we would have to compare them ourselves. That or NVidia and AMD are going to have to start their own certification program and publish their own specs so PSU manufacturers can put a "NVidia 4080 Certified" or "AMD 7800 Approved" sticker on the box to help tell consumers what to expect.
This is just another indication of just how far out of date ATX has become as a set of system building specs. ATX needs a major overhaul, but Intel isn't in a position to tell everyone what they have to do anymore.
The new ATX 3.0 standard will indeed require supporting 200% rated output for 100us. This should help quite a bit.
Just buy bigger psu , three times the power draw of pc the it will be enough .
@@kaiwenwu8148 that's not a bad rule of thumb. Maybe 2x instead of 3x, but at least 2x to be safe. Power supplies are typically in their peak efficiency around 50% loading anyway.
@@noname-gp6hk if account for 120% to 130% ocp trigger point of normal PSU, 2x of power rating should be ok.
So, a bit of a tipoff about testing for these transient spikes, as I recently had a problem with them myself that resulted in me just buying on on-sale EVGA 1300W so it can just soak up the problem, because thats far cheaper on time than any other solution.
The tipoff here is the BEST way I found of tripping it, happened literally within a few seconds every time, was fallout 4 VR. In general I was able to benchmark the system as I wished and never had even one shutdown during any benchmarks, however I had one shutdown while loading GTA online (weirdly), one shutdown in Phasmophobia VR, and a total of atleast 30 shutdowns in fallout 4 VR before I even got through character creation. I'd highly recommend it as the BEST way to see if a spike can take out your system power.
My System contains a 5950x, an MSI 3080ti, (64GB 3600MHz RAM) and was running on a Coolermaster 850W MWE V2, if anyone wants to know
3080ti isnt that far off from an 3090 ... afaik a 3090 can have spikes at almost 600w, worst case for a 5950x would be like 250w and for stuff like ssd, fans, lights etc ill give them ~50 to 75w ... espacially if you look at newer boards ... then i'll take 20% above it and this results in my psu capacity i want to buy for a rig.
Can you share the framerate you were hitting in Fallout 4 VR?
I feel like this has happened to me. I can't get my system to reboot without pulling the CMOS battery.
I had a similar problem like a decade ago with a 570GTX or whatever it was. Every time I looked at water in World of Warcraft the computer would shut down within seconds. Took a while to isolate the behavior down to that, but once I put realized that was happening I could reproduce it every time. Put a bigger PSU in the system and the problem went away. Transients are fun.
Bigger picture though, all PC components are ballooning in power consumption. Dennard scaling was already dying off for years, but it's really hit a wall in the past couple of years. A significant source of extra performance is now coming from simply letting components draw more power instead of getting more performance from a similar amount of power like we saw in years past. This is only going to get worse from all vendors as time goes on.
@@ngc4486diane People don't often understand, and there's a lot of bad advice out there. Pre-builts too often have substandard components, so I'm not surprised at all.
I've said it before and I'll say it again; I think Gamers Nexus is the true heir apparent to the legacy of The Tech Report. This type of thing is exactly the type of work we expected from TR back in the day with Scott, Geoff, Cyril, and co. If GN can figure out a way to make this kind of GPU power capture easy to accomplish I think it will have a similar effect to Scott's work with bringing frame time to the benchmarking game back in the day.
I really appreciate this experimental approach, rather than just presenting existing benchmarks.
Yours is the first technology channel I have $thanked.
This is extremely well done, Steve (and crew). A few of my colleagues have had this issue and it was tough for me to explain it to them what was happening with their systems. Now, I'm just going to refer them to this video, as it easy to understand the problem from a viewers perspective. Thanks for all the work you guys do over at GN!
Nvidia needs to install large capacitors on the input voltage rails for the upcoming RTX 40XX series as part of the reference design. This can be done quite cheaply.
A bit of a packaging challenge though - assuming that you allow the rail to sag to 9V, the minimum capacitance for the 3090 Master at 4:06 would be 20mF. Those caps are usually around 40mm tall.
@@lachlantheobald1819 Well, the 40 series cards are going to have monster heatsinks and will be more than 3.5 slots soo there is plenty of space to work something out.
4 slot cards here we come
@@ashc3765 Naaa we are just going to have the card with its own water cooled external housing and an add in card that plugs into the PCIE slot with dual Fibre Channel cables. The external housing comes with its own nuclear reactor you set up outside and plug a water hose into. That should make due until the RTX 6000 series that requires a antimatter reactor.
Nop, they actually need to up their efficiency instead of trying to simply powering up the gpu with an RTG dedicated for that.
GN, you guys are the true MVP's in the tech space. thank you SO MUCH for putting in so much effort for us to learn and understand PC's and tech. i've been having this exact issue with my 7900XT. transient spikes go way overboard and my SeaSonic 850w 80+Gold trips OCP and the system reboots. im in talks w/ the manufacturer now, but thank you GN for publishing these in depth walkthroughs as to the why and how. will likely get a beefier unit once all is said and done.
A fascinating and educational in depth discussion of a problem I was not even aware of. As an enthusiast builder, this kind of information is invaluable. Your decision to dive into more sophisticated testing and its associated educational benefits for amateurs like me is welcome and appreciated. I hope it works out for your continued success.
12:55 That moment when you start wondering if your 1000watt PSU will be enough. I actually got mine because my 2080TI and i9-9900k were giving me these exact OCP-ish shutdowns on my 850 so I figured I'd go overkill. So much for overkill... NVidia really should be building more capacitance into the card design if this is how their going to operate. It's ridiculous to expect PSUs to handle this kind of stuff as a general market feature.
Remember that throwing a bigger PSU at the problem can end up tanking your efficiency numbers.
PSUs are MUCH less efficient at lower percentage loads.
Going from a 1000W to 1600W PSU on the same system could end up costing you 2-4x the idle wattage.
This kind of thing is wasteful, and costly.
Maybe a PSU company could just make a capacitor bank for the GPU to sit in between the PSU and GPU
Strange, I ran a power OC bios (can't recall the power but it was the max available on any bios) on my 2080ti liquid cooled with a 215watt 8700k for years on a 650watt seasonic PSU with zero issues. Though I did need to upgrade as the 650 didnt handle my 3080
That might be what was happening to me with my 6800 and mild OCs.
@@kidShibuya Just like Steve mentions, the PSU brand and individual design matters as well. A high quality seasonic 650 could probably handle high transients much better than a poorly designed 850.
That's the biggest problem here -- you can't even go off your PSU's wattage rating, there's so many variables, including how well your motherboard regulated your CPU transients (VRM buffer caps could also have helped you there).
This is why ultimately these huge massive transients coming out of modern GPUs are such an issue. And it's one they could moderate themselves by providing better buffering on the GPU's power delivery, or not letting their cores push transients that high in the first place.
It's the randomness and multi-variable nature of this that makes it so troubling as a trend. Glad GN made this video, needs more general awareness.
Seems odd that the motherboard choice affects whether the problem shows up. Could be that some boards have extra capacitance on the 12V rail that "help out" the PSU when dealing with spikes. Or it could be that in some cases it's the motherboard that causes a shutdown if a 12V sag causes it to switch off the PSON signal.
could it possibly be the pcie slot has more throughput for voltage on some x boards?
Ge it can do 110w vs the "normal" 75w. so when it hits a transient spike, the slot compensates for it.
@@Dracossaint Seems like a legit possible theory, the motherboard should never do that though the spec is there for a reason. Above all though the GPU should never need to be drawing 700 watts and if it is going to be doing that the GPU should be the part with the extra capacitance to handle those spikes without requiring other components to carry the load for their poor design.
Both the amount of capacitance, and the amount of energy the inductors can store will affect the behaviour of a motherboard or GPU card during transients.
It’s not only capacitance. With transients this big, your board needs to be designed differently. If the copper carrying power to your PCIe slot is too thin, or it’s ground is not directly underneath, the added inductance can negate capacitance on the bus.
@@bdhale34 It's not the motherboard that should not do that, it's the graphics card's duty to not draw more than in total 75W (66W from 12V rail) from the board.
On the other hand many boards can provide more than 75W of power. I think since the RX480 there were no GPUs drawing more than 75W from the PCIe slot.
Thank you so much for this video, I have been unable to figure out why my PC has been shutting down, the way you broke this down incredibly well! A Digital Storm technician that stumbled across this video sent this to me, I'm hoping we can figure out a way to get my 4k computer back up and running reliably. I can't even game with the spontaneous shutdowns, with no consistency. Thanks again for taking the time to investigate this issue so thoroughly, definitely earned a sub!
25 years back when I was the dude that built all my friend's PCs I found out, that the most erratic and hard to reproduce errors could be attributed to the PSU. I had my list what things to go through if ppl were having troubles but it very often came down to cheap psu that were used in pre builts. Once the pc was upgraded the errors started to mount up. This seems to serve as a good explanation of these experiences. If you buy the PSU judging only by the specification numbers you can have a bad awakening. Thanks for you hard work
By what criteria should I buy a PSU then? Just on the price?
Besides all that you say and are at the Core of the problem, one should consider also that the majority of issues coinceiding with the PSU was also in parallel to 2 other factors,
One was poor Ground for PC chassis and the other one was also poor Grounded House Power plugs which also contributed in instability of the whole system..
that was happening in Older PCs (2012-2013 and prior) and Also the more mainstream PSUs were then of questionable Quality, more often than Today..
Ofcourse PC's weren't that demanding in Power back then .. and also Some brands of Quality existed.
I remember picking up a Totally Silent PSU (Passive Cooling, must have been 450W or 500W, dont rem actually ..that was at early 2000 and the price was then 135EU which was expensive at tge time) powering along a Ati 1600X.. its still runnimg today in a Cube setup Powering a Amd FX 55 System.
@@neutechevo poor electrical fuses were also often an issue but it was easy to rule out and just use a wall plug in a room with less load on the fuse. These issues often peaked during lan sessions as too many consumers were on one fuse. I didn't say that all old psus were bad. Just the ones oems cheaped out on in their prebuilts as it drove down their costs.
@@Dodara87 reviews i guess. Its hard to come by proper reviews for psus since dirkvader de shut down however. I try to reasearch as best as I can though.
@@benjaminolry5849 yes ofcourse. I said that more mainstream PSUs back then were of more questionable quality due to lower power demands.. also lots of us have more than one, burnt hdds psus mobos..etc.. and you could not pin point the actual problem.. i vividly rem. a problem for my own system.. and then i figured it out at sbout a month away was due to 1 or 2 screws of the motherboard not tight in the chassis.. these problems back them were more common.
The sheer exponential insanity that seems to be going on with GPU power consumption from generation to generation now is scary enough...
Don't worry, North America isn't getting any more than about 1500w on 1 circuit. So at least there is that limit 😅
@@ahfreebird Well your monitor consumes some power as well :D
Might get funny in the future.
@@ahfreebird ... my computer and monitors have to be on separate breakers (cause the monitors are enough to trip a 120V 15A breaker when added to my computer).
@@Ri_Shin_Marco No big deal, just gotta run an extension cord to an outlet on a different circuit.
Mainly with Nvidia GPU's. AMD is also rising a bit in power consumption, but their cards are literally hundreds 100+ watts behind Nvidias in consumption.
Also, both Intel and AMD are getting to kinda rediculous power draw on CPU's as well. Intle is already there, and AMD's AM4 socket maxed at 105w TDP's, the new AM5 Socket is going up to 175!
Thank you so much for doing this testing and making this video. Really appreciated.
Great video. Love this type of video. Thanks for taking the time and effort to do these videos.
Yeah got that first hand when I upgraded to a 3080ti. 850w should have been plenty for the steady state, but transients were enough to send the PSU into safety shutdown.
Since you asked for details Steve: EVGA 3080Ti FTW3 Ultra Gaming, ASUS MAXIMUS X Formula, 8086k Der8auer edition @5G all-core, Seasonic Prime TX 850 from 2018 couldn't hack it after upgrading from a 1080ti.
The 1kw model of the same line is doing fine, absurd as that is on a system that draws 450w average while gaming according to my UPS.
Great to see real user experiences like this to help establish that this is a factor to consider now!
@@GamersNexus not just that, even my cable extensions from asiahorse weren't able to handle my 3080, just because of the power not being clean enough
@@GamersNexus i have been running a gigabyte 3080 on a rm850 that is quite old for a good while and haven't seen these issues. It's a 320w hard limited model so I've made my own curve for it where it really doesn't do the stupid push past 2040mhz at like 1060mv. Mine mostly run 2025mhz at 1000mv so I wonder how this curve affects transients.. I'm not finished with the video yet obviously but I would be interested to see transients without Nvidia boost algorithm.
This explains an issue I had with a SeaSonic Prime Titanium 1000W about a year and a half ago. Shortly after adding a Gigabyte Waterforce 3080 to a Ryzen x570/5900X system, the PSU blew out. I could not understand how a PSU I had had for 2+ years across multiple platforms would all of a sudden give up the ghost, but to SeaSonic's credit, it only took itself out and every other part was fine, but I literally had no idea it could have been the GPU until now. Yikes!
Had my 3080ti kill a Seasonic Prime Titanium myself, currently in RMA.
Wow, GN - this is getting very professional. As an engineer I am very happy to see this "marriage" of quality knowledge and consumer/user friendliness. All the best to you. Keep it up :-)
the level of professionalism and quality of this video is amazing. GN it's not just your average tech reviewer that read tech stuffs on the product labels
A year later this information still rings true! Using some of this for some pc building for work. Thanks!
This honestly some of the best quality, and best explained tech video I have seen in quite some time! Very educational, thank you!
I am a CNC Machinist and shop owner. In dealing with old CNC equipment I have dealt with Controller failures from blown capacitors many times. (in the circuit to handle transient power draw) Thank you for this video explaining the problem to your viewers. (something I would just naturally assume when the system crashed) :)
Love your content. I'm surprised you haven't received a Netflix deal yet. Watching your take on Transients is way more entertaining than anything on streaming platforms.
I have a passing interest in all things computers, and I've learned so much from every one of your videos. Thank you so much for all your hard work!
20:45 And when we also account for the folding frequency, it might actually be better to build a separate setup that is offset by half a period, then combine the results into one function. That will eliminate any issues coming from the equipment generating errors due to the high resolution data processing.
22:10 - You can use mathworks MatLab for this, easily.
Thank you so much guys, for all the awesome content - loved this video
This is really excellent content! Thank you for doing this, and thank you for taking the time to figure out how to explain it clearly! That takes a lot of skill and effort!
THANK YOU.... THANK YOU .... THANK YOU !!!!
Since January when I got my 3080ti Hydrocopper have been fighting this issue and have never been able to resolve it. As mentioned, I DID NOT want to blame the GPU so I have spent countless hours trying isolate the issue of bluescreen shutdowns looking at everything else.
So, THANK YOU for helping me understand what the problem is and that I haven't gone insane !!!
This is excellent content - thank you so much for the work you're doing here.
I just had this problem. I upgraded my GPU from a 1080ti to a 2080ti, and in the process, I decided to change the case and water-cool the system, which also meant adding a lot of fans, a pump, and all of the RGB that comes with it.
The build that eventually had the issue was with an EVGA Supernova 750W G3, a GigaByte Aorus Elite Z390 MB (with a seated Intel 9700k), and a Gigabyte RTX 2080ti Waterforce GPU. The PSU was about 3 years old, and the system would shut down randomly after gaming for 5/10 minutes. Opening up MSI Afterburner and reducing the power-limit on the card to 90% solved the issue at a cost of performance, so I knew I had to upgrade the PSU. I wasn't in a place where upgrading the MB/CPU/RAM made sense - I'm waiting to see what this next generation of processors look like from AMD and Intel.
I went big and bought be quiet! Straight Power 11 Platinum 1200W. Needless to say, the system is stable now, even after overclocking the GPU and maxing its power consumption limits. If you're having this problem, try under-powering the gpu to see if you can stabalize the system. It's not a permanent solution, but it might allow you to game while you figure out your next steps.
This reminds me of when reviewers added "frame pacing" and 99th percentile frame rates to their review tests for GPU display capabilities. The averages weren't good enough to capture the user experience. One thing both types of manufacturers can do to help: add specs for "max transient voltage or current" over a specified duration such as 100 microseconds. GPUs guaranteeing that they won't exceed that value, PSUs guaranteeing that they can sustain that value.
I can see you amped the quality too (pun intended).
Really interesting find! Thank you!
this is brilliant. it's like watching a proper, high-quality tech investigative journalism. and HUGE kudos for slowing down the narrative a bit for old farts like me :D
Awesome to see this investigated and laid out in a straightforward way. IMO it seems like most of the blame lies w/ the GPU manufacturers for not controlling the spikes better, but I agree that overbuying on PSU is the easiest way for consumers to address it. I'm likely putting together a new build late this year. I was planning on once again reusing my Corsair HX1000 that's now over14 years old, but that might not be a good idea even though its 1000W... better to keep it in a system that won't have as high spikes and continue getting reliable operation from it and go with a newer unit for my new build.
I'd be interested in additional data on GN Extras!
In my experience a seasonic focus gold 1000W PSU is able to withstand a 3090 FE with a 5950x on a custom water loop. Before that one I had a Corsair RM850x that crashed during transients.
Awesome video as always. Side note, i see a lot of comments about csr audio systems. Didnt know my love for Bassy music was actually affecting the car in that way. Lol
Wasn't there an aftermarket overclocking GPU that came with an optional capacitor module you could stick directly into the backside of the GPU die? If memory serves, it was sometime in the DX11 era.
I feel like that was trying to solve this very issue for extreme overclocking.
I think 780 ti or 980
@@lgpa565 I remember now; it was the MSI Lightning line of extreme OC cards, starting with the HD 7970 and continuing with the GTX 780.
They called it the "gpu reactor".
Transits are a problem known for a good while. I suspect my old board was tripping because the caps were getting old enough to no longer deal with power fluctuations.
In the world today efficiency should be a huge concern, I was happy when AMD finally started to take it seriously after there older GPU's and CPU's. It's high time we start expecting more from new generation hardware, bigger and faster doesn't mean anything if it's pulling 2x to 4x the power of previous generations. Sloppy drivers and software play there part but with Gamers Nexus on the case we can finally hope for some truth and justice to be served. Thanks Steve and you're amazing crew, thank you as well for explaining how you're equipment works instead of just saying "it knows".
there is no choice, silicon based chips are extremely close to their limit, so they will get bigger and bigger and consume more and more .... and obviously they will be lot more expensive
efficiency doesn't matter. if you're so concerned about your power draw, you can always undervolt for a marginal penalty. more power usage an emotional, but not logical feels bad. so long as it can be cooled, it doesn't matter how much power a card uses.
@@zannare there is a choice, It’s just the more difficult one
@@zannare There is choice, just look to Apple and the M1 chips for a perfect example. As much rightful hate as apple gets for some of their practices, the M1 chips are insanely efficient. My m1 macbook has a battery that lasts two days easy, literally doesn't have fans and still doesn't get hotter than any laptop I've ever used before even under max loads.
It's definitely possible. Not easy, but there is a choice. The issue is the time and money that companies like amd and nvidia would have to invest, and there is littl incentive to do so if you can just offload the problem to the consumer by forcing them to buy bigger and bigger PSUs.
@@balls2thewall724 lolol efficiency always matters when it comes to lowering power draw. Were you joking?
Thank God someone is seriously talking about this. I remember when I thought getting the 980ti and 1080ti was a far cry from what was really possible for me. Oh how little I knew. And oh how much worse it can get, even from where we're at now. It sure feels like tripping people's breakers is the deep part of the iceberg but I wouldn't bet on that.
Thank you for all the effort doing your content. 👍
I haven't experienced any shutdowns. I have the 750W Super Flower Leadex III Gold , a 5800x3d with an Asus x570 Tuf board, and an Asus Tuf OC 3090. And thanks GN team for always providing insightful content.
good lord, the music at first had me thinking this wouldn't be the greatest of your pieces, but then you dialed it in and got all graphics-y... you guys produce the best content in the field. keep up the fantastic work!
Yeah I was afraid they went buzzfeed for a minute, then it got GOOD
Music over commentary still hurts the video imo but I'm willing to accept that I might be in the minority in this opinion
I’ve had this issue with my Strix 3090 and a 1000W Corsair psu…was very frustrating trying to figure it out. Keep up the great work!
I got a 3090 FE and a corsair 1000W power supply as well, luckily never ran into this issue even with some gpu and cpu overclocking.
@@Luke-yd4xq yeah I wasn’t that lucky unfortunately. I couldn’t figure it out honestly my friend helped my say for certain what it was
@@DairyQueen2187 damn, you got that with a 1000W PSU ? Do you have a lot of power draw from other sources ? Because if it's just the normal CPU + PSU + 1-3 SSD+HDD, the 1000W should be able to handle it no problem.
@@Winnetou17 Just a 5950X that wasn’t overclocked and my nvme drive. Don’t think the psu was faulty faulty but it was a serious problem
What psu did u get to fix it?
Easiest way to overcome this is for the 3 GPU Amigos to form a standard that PSUs would be able to meet and customers can choose them based on the 3 Amigos certified logo stuck on the side of the PSU case. Thus PSU companies do not need to fight to the bottom of the price war by eliminating components that would cope with the transients.
Thanks! Very interesting stuff and very informative.
Absolutely loved this video!
All the effort that went into this video really paid off. (I especially liked the animation 😅)
Hey Steve and GN team, what effects does Undervolting a GPU that exhibits high transients have?
Undervolting already has a ton of benefits, would love to see if it can be a way to deal with transients to avoid having to replace a power supply!
If the GPU itself can handle it I'm sure it will make a difference. You can also power limit the card and that will guarantee reliability
@@Mpdarkguy @Cătălin Power Limit wont fix as the max power limit of rtx3090 is ~350-400w depending on the model so it never should be able to even go to 600w+ spikes but lower voltage means there is really less power going into the card and therefore should improve the spike behaviour
I was thinking this as well. The down side is that the silicon lotto plays a role in how much you can undervolt, only allowing a generalized answer at best. It would still be handy to know if that indeed helps or not.
I undervolted my 3080ti FTW3 to 875mv which was quite a bit lower than the ridiculous stock current that EVGA decided to pump into this thing, (Which was absolutely unnecessary because I stay well above this card's boost clock and stable at 875mv) and while it did reduce my peak spikes, I still hit 425+ watts during spikes. (I originally would have them in the 460s) These spikes were just during regular gaming. I can't imagine what I still might manage to hit during some kind of GPU torture test. That said, I've never had a lockup or PSU shut down of any kind. Although, I am running an EVGA 750 Watt gold SuperNova, which are pretty rock solid.
So in total, I'm using an 8700k typically running at an almost 5GHz all core overclock, the 3080ti FTW3, two mechanical hard drives, three SSD drives, four 120mm Corsair LL series RGB fans, a single 120mm slim fan, a four port USB 3.0 expansion card, and about 8 USB devices total going into the PC with, typically, half of them drawing some kind of power at all times.
I'm probably OK because of the 8700k. I'd imagine if I put in any newer chip with 8+ cores, I might run into some issues.
Honestly just undervolt. It's so beneficial. My FE 3070 has it's power consumption cut by two, i can do 850mv at 1860mhz + 500Mhz vram OC stable.. On forza, i go from 205-220w to 110-135/140 at worst..
Great in depth video, well done
I started notice some of this back in 2005 while working on premium workstations with multiple hard drives. We started focus back then on more premium power supplies and raising the overhead on the psu. Errors in both initial and external variables in power affected the system far less with a better psu installed.
1: That music. Nice. 2: Back in the "old days", general wisdom was to get a power supply with a 25% higher capacity than what the system added up to. It was mostly to guard against efficiency drops, but it seems like it might be good to help with the spikes.
couldnt u put a bank of capacitors on the 12V rail?
This content is super informative and interesting, thank you guys for doing this piece.
Excellent information once again, thank you guys!! I’m already sharing this with friends who have aging PSUs and want a new GPU
I wouldn't rule out problems popping up even on Systems that ran well for years, since capacitors also age (tough not nearly as badly as they used to).
Also IIRC, when Furmark came up it used to kill GPUs and manufacturers then added some sort of "Furmark detection" which prevented the cards to go all out, but I don't know if this applies to current cards / Furmark versions.
You guys have again demonstrated why you’re one of the highest quality computer content creator on the Tubes. Great job raising the bar.
I haven't been following your channel for very long, but I really appreciate how detailed your analysis is! You're taking extra care to select a measuring device that's able to accurately and reproducibly capture the issue and you have taken extra steps to account for potential sources of error in every piece of the measuring setup. The trigger delay of the clamps is something I didn't even think of, so I was pleasantly surprised that you mentioned it and explained how you were going to mitigate it.
You had me at "retroactive warranty". Class act, these guys.
Job done is insanely huge.
So much respect to you guys. You are awesome
Now THIS is tech journalism, I really hope you guys can keep on doing this kind of content for as long as you are around.
Well, except for the first 5 minutes where I kept waiting for Chris Hansen to bust in the door o.0
So few people out there actually understand and take the time to learn about hardware, and then really point out flaws. Not just surface-level or consumer obvious ones, but really deep dives. When you guys speak, people listen! Your channel makes a big splash.
Every generation the power draw grows, the heat grows, and the heatsinks get bigger and bigger. It's not sustainable.
That last part isn't really accurate. Power consumption (for both GPUs and CPUs) has see-sawed over time, but definitely with the 40XX series we're looking at record-setting levels of power consumption. With this generation, Nvidia is trying to beat AMD by throwing efficiency to the wind and just pouring raw power through GPUs to force faster clock speeds. This video shows one of the challenges of doing that, in that if your PSU draw 450 watts normally, it might need to be able to survive transients of 1500 watts. Both the day-to-day operating costs and the cost of requiring such a robust PSU will make these GPUs very poor value propositions.
Also, by taking this irresponsible approach of "MOAR POWAH!!!!11!!" and boasting about their power consumption numbers, they're just asking governments to step in and enact hard and fast regulations limiting maximum power consumption. And that likely will just make everybody unhappy.
It might not be sustainable, but it doesn't have to be. At least for mainstream consumer cards. Consider what the current high end cards can do: 4k, 120FPS, ray tracing. At least two, possibly all three. Whats left on the graphics wish list? 8k? Impractical on account of the shear size and cost. 4k VR? Already have it, just turn off the RT. More FPS? Diminishing returns past 60 already, and your not running competitive FPS like CSGO with 8k textures and ray tracing. Therefore, hello 600FPS.
There are just not any technical feature left to add in, and while there are a few niche combinations of games that can max out a 3090 and have sub 60FPS, lets say the 4090 doubles the graphics power for not double the power cost. RT is the final word when it comes to graphics features (I'm saving about 2 pages of explanation with this) and that has a more or less fixed compute cost. So lets say the 4090 has enough graphics power to run anything that you can reasonably run. Someone will have to draw the line in the sand and say" Our card can do 4k VR with RT with 8k textures with 240FPS... what do you not have phones, err 8k VR, err eyes that can break physical limits?.
So the 50 cards will be refining the design, dropping power, and dropping cost.
Thats not accounting for the artistic/scientific types, but they are going to always be wondering how they can fit xx90 #5 into there build.
@@megachonk9440 considering how relatively small the market for high-end enthusiast graphics cards is, government regulation vis a vis power consumption seems highly unlikely.
@@sharpe3698 Given how the trends are, we might be seeing silly power consuming on low and mid tier graphic card where power consumption vs performance become a cost vs value issue.
@@nickierv13 As long as cards *can* advance, games will continue to push the boundaries of what's possible. Everyone always says that there will be a limit, but it's never been true. There will always be some level of detail, some additional effect that opens the floodgates to more performance being needed. Todays that's raytracing, tomorrow it might be in-game fluid simulations, or per-hair physics or SOMETHING. No one is ever satisfied with stopping and saying "good enough".
Amazing stuff, thanks guys!
Very informative, great work ! That explains what I have been experiencing: I have a 550W power supply which should be enough for my 3060Ti and i3 10100 but every once in a while I get shutdowns like this.
Would have been great to see what happens in lower-tier cards, like a 3060 or 3070. Sure, the 3060's TDP is 170W so most PSUs could take a transient as normal, but consider someone on an older 450W power supply, which technically meets Nvidia's minimum spec.
Im using an rog strix 550w with a 3070ti , not even min specs lmao
@@blindshot709 is it... working?
@@KellonMelon65 yes... 6 months no shut downs
@@blindshot709 it is probably throttling
@@Metalwrath2 how would it be throttling when it never goes above 350w leaving 200w head room for spikes . also that 350 is under full load lmao gotta love 65w cpu's
Should also note that certain models of power supply are more vulnerable to transients than others. A 750W EVGA G6 might be able to handle a 3080 just fine, while a 750W EVGA GA likely will not. The Seasonic Prime-based units are also problematic in this regard, just to name a couple of the more popular lines. Either way, GN did a fantastic job of explaining transients here, and I look forward to what's in store.
Also, I suggest looking at Aris' Powenetics V2 which he announced recently; could be pretty useful in testing.
It basically depends on what BIG capacitor is used within power supply, so it can withstand sudden spike
@@DimkaTsv There is an awful lot more to transient response than just how big the capacitors are. Capacitance is an important part of it, but you can't know about whether a PSU can handle a 3090 just by looking at the caps.
@@nunyabeeswax3012 Sure, i guess... I am no electrician in any way, so just answered based on own knowledge, which is limited