We don’t often talk about power supplies, but Intel’s new ATX12VO spec—that’s an ‘O’ for ‘Oscar,’ not a zero—will start appearing soon in pre-built PCs from OEMs and system integrators, and it represents a major change in PSU design.
The ATX12VO spec removes voltage rails from the power supply, all in a bid to improve efficiency standards on the PC and meet stringent government regulations. But while the spec essentially removes +3.3-volt, +5-volt and -12-volt and +5-volt standby power from the PSU, they aren’t going away—they’re just moving to the motherboard. That’s the other big change, so keep reading to find out more.
Power supplies are definitely one of the more cumbersome parts of a modern PC build, so any changes there can potentially have a big impact. The new Mac Pro has really shown how a modern PC can be designed to not use ugly and annoying cabling, opting instead for various pogo pins and properly aligned connectors.
Sure, that would be much harder to accomplish in the open ecosystem of PCs, but for an easier building experience and thus potential access to a larger segment of the market, players in the PC industry would do well to come together and take a long, hard look at the Mac Pro and how to replicate some of its innovations into the wider PC industry.
Well, I’ve got a mixed reaction.
On the one hand, it will clearly make the power supply more efficient, but I’m concerned about making the motherboard less efficient. Modern computers use 12V lines with regulators on the motherboard to buck the voltages to those required by the motherboard and CPU components. These modules are called VRMs:
https://www.gamersnexus.net/guides/1229-anatomy-of-a-motherboard-what-is-a-vrm-mosfet?showall=1
These tend to run hot already, even prior to this new spec. I’m concerned what adding even more VRMs to the motherboard will do. Certainly we can/should get rid of the voltage lines that are never used, but in terms of peripherals, I really question whether the motherboard can do this more efficiently than the power supply.
My workstation uses a 1000W 80+ gold power supply, while I don’t have a way to measure it’s actual efficiency, it runs cold and the fan never kicks in. I’ve over-provisioned it for stability and in case I were to buy a second GPU, the max power I actually use is ~500W under load. The motherboard on the other hand does produce heat even at idle. Everything on the motherboard produces heat, but the VRMs in particular are notably hot.
So here’s my question/concern: is this actually solving anything or is it just going to shift the burden of powering peripherals to the motherboard to incur power conversion losses downstream? Unless there’s a clear net benefit overall, then I’m not so sure I want to add more stress and heat on the motherboard just so power supplies can claim better power efficiency – it’s somewhat of an artificial win.
Maybe peripherals themselves should provide their own buck regulators to produce whatever voltages they want from the 12v rails or to avoid power conversion all together. This would minimize current requirements and distribute heat away from the motherboard.
Incidentally, my gigabyte BIOS has a nice breakdown of temperatures and fans for the motherboard, and I’d like to graph the heat output of VRMs across loads, but unfortunately the it8686e chip isn’t supported under linux, and it doesn’t look like any progress is being made on this front…
https://github.com/lm-sensors/lm-sensors/issues/167
Therefor lmsensors only gives me ACPI and coretemp, which limits my ability to monitor temps and control fans on the motherboard. Is there any chance that anyone knows what ACPI temperatures map to on gigabyte z370 motherboards?
I can’t find documentation for this anywhere. I assume the first is an ambient temperature because of how low it is. Can anyone recommend a Z390 motherboard that works flawlessly with linux monitoring & fan control? Something with remote AMT would be desirable too. The lack of monitoring & fan control has been a lingering issue. It’s only minor, but still frustrating.
Alfman,
You probably know it, but the curve of efficiency of PSUs are related to the power effectively consumed, if your system load is way less than the 1000 W it is labeled, you are not getting the maxim efficiency it can provide. We should really look for what the system needs before buying PSUs, but, alas, for what I have read, unless only a fraction of what it can give is used, the difference of efficiency is really small, though the difference on $ is big.
Most problems associated with motherboards come from 2 components: capacitors (including those used in VRMs) and VRMs. By my experience, a good PSU outlives greatly motherboards so, I’m not sure it will be a good trade off.
Also, it should be noted that there is a lot of experience on providing just one voltage to motherboards, that is exactly what we get on notebooks and their “brick” power supplies, even though, the wattage is way less than what we usually need on desktops. For what I have read, most bricks have an efficiency of equal to or less than 60%.
So, I’m afraid we may end up getting PSUs with efficiency close to that of notebooks and the same low efficiency we already suffer from VRMs. Only time will tell.
acobar,
Yeah, I don’t know the power curve for my specific power supply, but in general the peak efficiency seems to be around 50%.
https://www.hardwaresecrets.com/understanding-the-80-plus-certification/
https://www.tomshardware.com/reviews/psu-buying-guide,2916-3.html
I could have used a smaller power supply, 500W would probably be the minimum for my current system. But I wanted to be able to add another GPU if needed. Also this is a separate issue, but I’ve found that low wattage power supplies have much lower “hold times” because they have less capacitance. Naturally a 1000W power supply needs to have enough capacitance to provide 1000 watts during the AC zero voltage crossings, but if you only use 500W you essentially double the hold time out of the same components. The reason I care about this is because the power used to go out very frequently and on several occasions my 3 computers on 3 separate UPSs would shut down between the time the power goes out and the UPS kicks in. One computer was effected maybe 10% of the time, another 2% of the time, and my ARM computer I used for security cameras was ~50% of the time. The ARM computer uses very little power, but even a tiny dip like running a heavy appliance and it would reboot. Obviously I could upgrade my UPSs to always online models to guarantee no gap at all, but this is less efficient than a passive model that stays out of the loop until the power fails. Anyways long story short: I solved the problem in all cases with over provisioned power supplies. Maybe there’s a better solution, I’m all ears, but until I find it I know this just works.
My ARM SBC mentioned above used a power brick. Most laptop power supplies I’ve seen use 19+V. I think it’s easier to run higher voltages to the laptop and then regulate them inside. Otherwise the laptop power connector would have to have to support multiple voltages at higher currents, which would be kind of cumbersome.
Many of the voltage buck converters you find for electronic circuits are 80-90% efficiency range, but the problem is when you use a buck converted behind another buck converter you have to multiply the efficiencies: 120V -> 12V = 85% and 12V -> 5V = 85%. Therefor the converters in series are only 72% efficient whereas you might have had 85% efficiency for 120V->5V in the first place. Anyways you are right that time will tell.
Hum, I will not ask what is your UPS brand, but it should have enough power left to do the switching, as the producer knows the circuit reaction time and the labeled wattage is also known, shouldn’t it? Are you sure the combined load does not exceed the UPS spec?
One problem I had, was with surge and lack of grounding, the UPS couldn’t cope with this situation and the solution was to properly ground it.
acobar,
Yes, my main servers (including that ARM computer) are on one cyberpower 1500VA UPS and my desktop is on another, neither are overloaded and the other attached equipment does not reboot. I use other UPSs for network infrastructure…it’s a bit much, haha. Anyways I’m quite confident it was an issue with PSU hold up time.
https://blog.fsp-group.com/en/the-importance-of-hold-up-time/
Only some computers were effected, while others had never experienced an issue at all even on the same UPS. In all cases, the power issues have been fixed by upgrading the power supplies.
There’s another factor that I didn’t bother mentioning before now, which is simulated sine wave versus real sine wave. My newest UPS has a real sine wave, but they’re not all there yet. This can theoretically cause problems with certain power supplies too. I have already tested all my computers on simulated sine wave and they ran fine. It just seems that some have trouble staying powered on during the gap in a switchover event.
You know, it probably would be hugely informative to test my UPSs to see exactly how long they take to switch over. I’ve also got some solid state relays I could use to test exactly how long and under what kind of loads the PSUs will sustain power before shutting off. This would be the ultimate test, but I’m just not feeling the motivation right now, haha.
That’s curious. I’ve wondered about the safety of running multiple computers/devices on different UPSes disconnected from a common ground. Ethernet devices have galvanic isolation, but not all busses do and I’m not sure if different voltage potentials could risk burning them out.
To me, that sounds like what APC was competing on back in the 90s when, as a geeky aspie kid, I was calling 1-800 numbers and ordering product catalogues: “Standby UPS” vs “Online UPS”.
In a standby UPS (the cheaper kind), the battery side only kicks in when needed and capacitors are used to bridge the interval before it kicks in… capacitors which can be sucked dry by the house side of the circuit in a real power failure.
In an online UPS like the ones APC was marketing, the circuit is like what you’d find in a lot of battery-powered devices, where the battery is always online and it’s just a question of whether or not it’s getting recharged by AC power at the moment. Online UPSes only allow power interruptions if they’re overloaded or the batteries need replacement.
> and meet stringent government regulations
I look at that and think is it nothing more than a bad idea that will occomplish nothing of the good intentions.
Government: Power Supplies must be efficient.
PSU Makers: OK we can do better at one voltage (but others are needed and conversion on a motherboard will be less efficient, not to mention than PSU prices will not come down but motherboard costs will go up (motherboard prices have already got f’ing riduculous, it’s like choose one with onboard gfx capabilities you will never use to lower the price!).
I guess with only one spec to refer wattage to though, that the many poor PSU’s will go away, so that is a positive. (though ass quality componants will still be a thing!).
I think that due to the law of unintended consequences it will achive nothing of what is the intention and it will make things more expensive for people. (kind of like the plastic is evil crowd, where something non plastic is better even if it is worse for the environment)..
It isn’t “Power supplies need to be more efficient,’ it’s “desktop computers need to be more efficient.”
Individual computer components aren’t what’s being regulated.
Carrot007,
Well, I’m not anti-progress. The ATX standard is old, and sometimes the industry needs a kick in the rear to do a better job. Green energy standards in the past have certainly helped make electronics (and vehicles, etc) more efficient. Realistically these advances probably would not have come from manufacturers acting in their own interests. So I get the need to shake things up, however what good does it do for power supply manufactures to pass the buck? They may get credit for reducing power supply inefficiencies and reducing their costs, but it may not do much to improve energy efficiency overall I don’t want to see this just add cost/complexity/inefficiency elsewhere.
I indicated earlier that I’d be ok with dropping rarely used voltages, but for external peripherals that still use 5V, Connecting regulators in series (as this standard implies) could potentially increase inefficiencies. I just don’t see how a motherboard is going to do a better job than the power supply. Is there hard data that shows this will provide a net benefit?
Good catch.
This smells like them trying to move components with a higher failure rate out of an easily replaceable commodity module, and onto a component where replacing it involves more Intel chips being sold (either on the board or in conjunction with a CPU upgrade)
I was recently looking at PSUs. Most options I’ve considered have a 10 or 12 YEAR warranty. This outlives other components in my stationary computer. There is no higher failure rates, I’m most likely to replace motherboard before PSU dies.
I’ve been looking to replace PSU in order to reduce amount of heat generated. My current PSU is 9 years old, it still functions good, but the efficiency isn’t great. I’m going to replace it with 92% efficient Seasonic.
zdzichu,
This is probably just a case of very bad luck (many years ago at that), but I was upgrading an old PC with a new seasonic power supply and it began smoking. I immediately unplugged it but the damage was done and it fried my computer, They asked me to send in the PSU and damaged computer parts. They immediately acknowledged the PSU was defective, but they wouldn’t replace the computer outright, it took a lot of prodding but they finally sent a check for about half the market value. So while I got more than I paid for the PSU, it was a net loss for me. On the plus side, I actually think seasonic deserves credit for at least trying to work with me, some companies are just awful about honoring warranties after they have your money.
I’m glad you’re seeing decent kit, and that you have that option. I’d suggest that you even looking at PSUs yourself means you’re in the minority of users however.
There’s an awful lot of generic power supplies out there, and there’s no reason to assume that the components on a low end motherboard will be any better than those in a cheap power supply.
Warranties to consumers are usually on the machine as a whole for a couple of years, not on individual components.
Power supply failures are one of the most common hardware issues with PCs.
If half the PSU is moved to the motherboard, and those parts blow, then the entire motherboard has to be replaced. For Intel boards, that means new chipsets, and if the board is older it means a new cpu + board. That means intel ultimately profits from an otherwise unrelated failure.
Ever heard about PicoPSU, someone ?
Okay, you made me do the work so…it’s one maker’s 120W supply one feeds with 12V input and might be just the
thing and sits there above the ATX socket that’s half the thing’s size, dissipating power, generating 3.3V and 5V rails….
Next, you’re going to make us laugh showing the skinny arse 12VO supplies, and finding us the spec. for 12VO that talks about how it is one attaches 13.7V LiPo storage as a UPS to it (safely,) and gets all kinds of powerline, factor, and UPS status info over ACPI interfaces.