It’s been six years since Intel came up with ATX12VO PSUs as and alternative to ATX power supplies, replacing PC’s multiple voltage rails with a single 12V rail (hence the name: “12V Only”) to improve energy efficiency. So far, it hasn’t gained much traction. Intel is now introducing a new “v3” version, which brings deeper changes. They’re quite interesting and could justify a transition, but there’s also one big caveat.
ATX12VO v3 power supplies will probably be among the innovations presented by various manufacturers at Computex 2026 in this week. Whether Intel itself will specifically showcase this technology is uncertain, because it is not exactly something that interests mainstream consumers directly, like a GPU would. However, documents from a presentation about the technology, apparently intended mainly for PC manufacturers, have leaked online and revealed what this new version is about.
Goodbye Standby
ATX12VO v3 power supplies will probably not be compatible with previous versions (and they are also incompatible with ATX). Connectors will change, but also the power delivery itself. The most important aspect is that the “Standby” rail disappears from these power supplies. Thirty years ago (time really flies…) this was one of the biggest innovations ATX power supplies introduced compared to the historical AT PSU design. It meant that the power supply had to remain partially active at all times in order to power components that are always on, such as the real-time clock and CMOS memory storing firmware settings. It also enables features such as Wake on LAN feature.
The original ATX12VO power supply technology also included a Standby rail (12V instead of 5V as used by ATX), but ATX12VO v3 removes it. This should save energy because this functionality constantly draws power from the electrical grid even when the computer and PSU are switched off. It is not yet completely clear whether ATX12VO v3 power supplies and PCs using them will have truly zero power draw when powered off (as old AT computers once did), but at minimum Standby power consumption could be reduced.
If a computer requires functions such as Wake On LAN, these should still remain supported. ATX12VO v3 power supplies introduce two activity states—High Power Mode and Low Power Mode. The latter still keeps 12V power rail active. We assume this could be used in combination with modes such as Modern Standby.
Finally PSU monitoring—but also active overload protection
A very significant—and in some respects perhaps the most interesting—innovation is that these power supplies finally gain some form of communication with the motherboard. All ATX power supplies currently function as a “black box” from the perspective of the rest of the computer. They are one of the few components that a PC cannot monitor in any meaningful way—you cannot read temperature sensor data, fan speeds (to verify that the fan is operational), or similar information, despite such telemetry potentially being very useful. If the PSU could provide telemetry, its electronics could report total system power draw without requiring an external wattmeter—which may itself be inaccurate—or warn about problems.
ATX12VO introduces such capability, for the first time. It is not mandatory, but power supplies can implement it by adding four extra pins to the motherboard power connector (bringing the total connector size to 12 pins—these additional pins are placed on a detachable section of the connector so compatibility with the 8-pin version remains preserved). Pins 9 through 12 are dedicated to the PMBUS communication interface. This is an already existing technology used for similar purposes in server PSUs, meaning support infrastructure already exists and extending it to desktop PCs should be relatively easy.
One of the most important improvements PMBUS enables is overload protection. The problem with power supplies today is that they have a certain capacity you should not exceed, but users generally have little idea how much power their computer actually consumes or whether they are unknowingly overloading their PSU.
Through PMBUS, however, the PSU can finally provide feedback—specifically via an I_PSU% signal, which should represent PSU utilization percentage. This allows the PSU to warn the motherboard when load exceeds 100% of rated capacity. The computer (through the processor and motherboard’s firmware) can then respond by reducing power consumption of the components (CPU and GPU mainly) until load returns to safe levels.
Ideally, PCs should have supported something like this long ago. Today, at best, you have overload protection baked-in inside the PSU itself—but those features can only respond to load exceeding the PSU’s capability by forcing and immediate shutdown. The PC powers off without warning and without allowing the OS to prepare, making it similarly problematic to a sudden system freeze or power outage. You lose unsaved work, your game crashes, and in worse cases this can result in corrupted file systems or damaged data on SSDs.
An ATX12VO v3 power supply implementing PMBUS would solve this elegantly and correctly by allowing the PC to slow itself down slightly without stopping operation entirely. This is probably the single biggest advantage and we think this by itself could justify moving to transition desktop PCs to this power supply standard.
…but with connectors reminiscent of Nvidia’s melting 12V-2×6
The biggest downside, on the other hand: the physical connectors themselves will also change, becoming smaller and potentially more fragile. The pitch of the pins and their protective structures on both motherboard and CPU power connectors shrink from 4.2 mm to 3.0 mm, reducing overall connector size and making them easier to fit onto mATX and especially Mini-ITX motherboards. Unfortunately, these new connectors appear to use the same connector family (which Molex calls Micro-Fit) as the 12+4-pin connectors adopted for GPU power delivery—which uses the exact same 3 mm pitch—which could mean identical physical and electrical characteristics (and possibly equal reliability issues) of the pins and sockets. These connectors, however, will use eight-pin versions for CPU power and either eight or twelve-pin versions for motherboard power (the 12-pin version is required to support PMBUS telemetry).
On graphics cards, 12+4-pin GPU connectors fail dangerously with worrying frequency because they are not robust enough and can overheat so much that they destroy cables and connectors on power supplies and graphics cards. Cases have even been recorded where the cable caught fire, which is a completely unacceptable risk. Let us hope that Intel has learned from this shameful fiasco and has somehow mitigated this danger.
- Read more: Uneven Resistance and Lack of Headroom: Why 12V-2x6s Fail
- Read more: 12+4pin connector still less reliable than 8pin. At as low as 300 W
However, it remains questionable whether this connector type will be mechanically more robust when used in ATX12VO v3 power supplies. At minimum, we hope Intel will have enough common sense to preserve sufficient safety margins—which Nvidia neglected with 12V-2×6 connectors.
This casts a rather unpleasant shadow over what is otherwise an interesting technology—the risk of burned connectors would no longer be limited to graphics cards and PSUs, but would additionally introduce potential damage risks for motherboards themselves (and in two separate locations). In theory these connectors could still be swapped to larger, more robust ATX-style connectors, but this would obviously sacrifice physical compatibility and require adapters with the Micro-Fit based v3 standard. However, if PSUs, motherboards, and complete systems utilising ATX12VO v3 are already being prepared now, it is probably too late to change connector designs.
Significantly lower power consumption both idle and under load
Overall, ATX12VO v3 power supplies are supposed to reduce desktop PC power consumption both at idle and under load. Intel provides reference measurements comparing systems using a Core i5-14400F processor, GeForce RTX 4060, 33 GB of RAM, 1 TB NVMe SSD, and 500W power supplies. According to Intel’s measurements, a system using an Asus Z790M-Plus Prime motherboard with a Be Quiet! E10-CM-500W PSU consumed 29% more power during idle and low-load scenarios (Energy Star Idle Mode and Energy Star Measured TEC tests) compared to an ATX12Vo v3 system.
However, improvements were reportedly also measured under load. The ATX power supply demonstrated 12% higher power consumption in testing (for example Cinebench R24, Geekbench 6 ST/MT, and gaming benchmarks using Far Cry 6) compared to ATX12VO v3. Measurements for the ATX12VO v3 configuration were taken using a Lenovo LOQ gaming desktop equipped with an AcBel PCK010 500W power supply. It should be noted that its motherboard also necessarily differs from the Asus board mentioned previously, which cannot be fully separated from the measured results.
These improvements in efficiency and lower power consumption during idle and even while powered off should help satisfy various certification programs and regulations imposing stricter requirements on PC power consumption. As a result, these power supplies could be adopted by major PC manufacturers that need to comply with such regulations.
The bigger question is whether ATX12VO v3 power supplies and motherboards will ever appear outside OEM systems and enter the DIY market, where users purchase components separately and build PCs themselves. Previous ATX12VO versions never achieved meaningful adoption there. However, version v3 introduces several improvements—especially the PMBUS capability—which might finally make it worthwhile, although abandoning decades-long PSU and motherboard compatibility is obviously unfortunate. That’s assuming the reduced-size connectors do not prove as unreliable as GPU 12V-2×6 connectors, though.
Although ATX12VO v3 introduces these new connectors and telemetry functionality, there remains a theoretical possibility that limited compatibility with older ATX power supplies could still exist through passive adapters or conversion cables. Previous ATX12VO versions faced difficulties due to different Standby voltages (ATX has 5V Standby rail). Ironically, removing the Standby rail might actually allow some compatibility with traditional ATX PSUs, albeit without the telemetry features and Low Power Mode functionality. Whether computers could actually function correctly under these circumstances remains uncertain. Documents that leaked online mention nothing about such partial compatibility with ATX, suggesting Intel is probably not explicitly targeting this scenario.
Sources: Momomo_us, techPowerUp
English translation and edit by Jozef Dudáš
⠀
a much less bulky connector is critically needed,something earlier 12VO failed to provide
but besides that we also need sane power limits and efficiency optimization, components are so heavily overclocked out of the box you can cut their power in half barely touching performance, what’s the point?
The advantage, as I see it, is that users don’t have to manually overclock in order to squeeze maximum performance out of the processor—at least until they eventually hit the cooling limit, unless the user changes settings manually or unless that limit is already configured differently from the factory, for example because of weaker VRMs. That limit is, of course, ultimately determined by the cooler design itself. 🙂
the only advantage is getting a slight edge in benchmarks
some extra oomph is fine, but getting 10% of performance at the cost doubling the power is not
Sure, energy efficiency decreases as you push toward the highest frequencies. 🙂
“sane power limits” that is a problem, if you need more computing which requires more power (like HEDT CPUs and several professional GPUs) you have a serious problem due to limitation, not to mention people that use their hardware for OCing and benchmarking. What I think companies like Intel, AMD and Nvidia should do is push the OEMs to add a selection pop-up feature when first installing the hardware (motherboard or GPU). Whether the user would like to run his hardware at the absolute limit which will give him 10-20% better performance, or run the hardware at more efficient curve cutting the power by 30-50% which will also make it quieter and cooler, then just load the bios settings and the user is ready to enjoy the hardware however he prefers. I think that choice is always better than artificial limitation, I know many that would prefer to run their hardware one way, and many that would prefer to run it the other way. There is no right or wrong, just preferences.
in the olden days when you wanted to push your hardware hard you overclocked it, but for the past decade or so it’s already overclocked out of the box (sometimes limiting total power, but still allowing lesser loads to push clocks to the very boundary of stability)
you may say it’s just preference, but defaulting to that makes PSU bulkier, mobo more complex and expensive, graphic cards bigger and more expensive, this trend causes a chain reaction
in the past you bought the more robust solution only if you wanted to OC, nowadays mobo needs to run all processors designed for the socket, including ones pulling 300W in peak
and sure, HEDT have to run on higher limits, but thanks to higher surface area cooling them is easier, and generally they are much more efficient as their clocks are lower anyway
Thank you for the information (do you have any sources supporting your claims?). Well, we’ll see how things turn out in practice. 🙂
I’m not sure what exactly you are asking about, but except some mITX boards that explicitly limit CPU support to 65W TDP parts (and sometimes 35 instead) all boards have to handle the entire lineup, including 250W TDP and likely over 500W transient loads
my H61 and H81 boards were barebones, even no heatsinks on VRM, and yet they were entirely capable of handling non-overclocked flagships of their generations, nowadays the base chipset boards are ghosts and the cheapest boards available have 3x more phases and massive heatsinks
PSU? try finding one below 500W, which is troublesome even when you tune your machine as the idle load falls so low the PSU drastically loses efficiency, and even under load it’ll land in like 30% load while you’d rather have it at 50% on sustained load and have some reserve for spikes
graphic cards yield VRMs even more complex than mobos, including mixing power supply through the slot and plugs, enormous coolers (usually with mediocre fans) while the GPU configuration could easily run on 70W alone if clocks and voltages were set smarter
I’m not saying the products of the past were perfectly optimized, but at the very least they weren’t pushed as hard
nowadays “overclocking” your CPU means making it run on the same boost clock with more threads loaded as adding even 5% higher clocks is beyond the capabilities of conventional cooling and power delivery
and I know I can tune things down, but even when I do so I still have to pay for all these overbuilt components
Ok, thank you for the information and exhaustive comment. That’s sufficient for me. 🙂
But these predefined profiles with different power limits that you describe already exist for processors and graphics cards, don’t they? In the latter case (graphics cards), there are often even two different BIOS versions specifically so users can quickly switch between prioritizing higher performance or lower noise. For motherboards and processors, it’s even more detailed. Besides setting power limits, there are quick presets covering various maximums—from genuinely power-efficient options all the way to unlimited settings, where the limiting factor tends to become the cooling performance of the CPU cooler itself. 🙂
the issue comes from defaults being overblown forcing more complex physical designs
RTX4070Ti chip can be tined down to 70W and remain majorly powerful as it’s moving very close to the optimal setup, but the external power cord is still required to boot the card, which doesn’t make much sense
on the other hand… you can potentially buy a pro card at 4-5x the price and then replace the cooler… which isn’t ideal either
Yes, there are more options and it’s up to each one of us which to choose 🙂
ATX12VO was made from the start for prebuilts and OEM machines, where manufacturers already use their own solution to meet regulations. While I like the changes, I know it most likely won’t come for DIY market (other maybe entry level products that are aimed for SI).