ATX 3.1 aligns the PSU spec to PCIe CEM 5.1. The 12V-2x6 connector replaces 12VHPWR, carries 600 W, and reports available power on two sense pins. Size for transient excursions, not average wattage, and prefer a native 12V-2x6 cable over an adapter.
At a glance
- Class
- PC Hardware
Provenance
- Nov 16, 2026Published
TL;DR · ATX 3.1 power supplies
- The headline change is a connector, not a wattage. ATX 3.1 aligns the PSU spec to PCIe CEM 5.1 and replaces 12VHPWR with 12V-2x6. Intel says new designs should mount only the new connector.
- Check the cable, not the badge. The two sense pins in a 12V-2x6 connector tell the card how much power it may draw. A PSU that configures those pins properly is worth more than a higher wattage number.
- Size for transients. A single add-in card may now be rated to 675 W and is allowed brief excursions above its sustained power. Average wattage is the wrong input.
- The 3.1 badge says nothing about efficiency. The required floor is 70% at full load. 80 PLUS and Cybenetics ETA are separate voluntary certifications, and they are the ones worth reading.
Most of what gets sold as “ATX 3.1” in 2026 is a rebadged power supply with a new cable in the box. That is not a scam, because the revision really is mostly about the cable, but it does mean the version number on the box tells you far less than the spec sheet does. The things that changed are specific, they are documented, and they are worth knowing before you spend money on a unit you will keep for a decade.
Here is what Intel’s own document says changed, what it means when you plug a GPU in, and the two timing numbers that decide whether your machine survives a power flicker.
What ATX 3.1 actually changed
Intel’s design guide states its scope in one line: “This document aligns to ATX Specification Version 3.1.” The revision exists to bring the power supply specification in line with the PCIe CEM 5.1 add-in card specification, and it lists four substantive changes (Intel, doc 336521 rev 2.1a):
- Two new auxiliary power connectors at 600 W. The 12V-2x6 connector carries 600 W on the 12 V rail, and the 48VHPWR connector carries 600 W on a 48 V rail. Each has four sideband signal conductors so the card and PSU can negotiate configuration directly.
- A power excursion allowance. Cards may make brief high-current demands while keeping average power inside the specified limit.
- A higher single-card ceiling. The maximum power for a single PCIe add-in card rose from 300 W to 675 W.
- An optional 48 V rail. Intel expects it mainly in data centres and does not require PSU support for it.
The fourth item is the one to ignore for a desktop. The first three are the ones that change what you should buy.
What did not change is just as important. The document’s efficiency table still sets the required minimum at 70% at full load, 72% at typical (50%) load and 65% at light (20%) load. The low-load requirement is 60% at 10 W or 2% of rated output. If you assumed a 3.1 unit was inherently more efficient than a 3.0 unit, the specification does not say that.
12V-2x6, the sense pins, and why adapters are the weak link
The 12VHPWR connector introduced in PCIe CEM 5.0 has been deprecated. Intel’s wording is unusually direct: the 12V-2x6 connector “replaces the 12VHPWR connector introduced in PCIe CEM 5.0, to address reliability concerns that were observed in the 12VHPWR connector. New power supply designs should mount only the 12V-2x6 connector and the 12VHPWR connector should be deprecated.”
The two connectors are close relatives, and Intel describes the 12V-2x6 as mechanically identical in most respects. Two changes matter in practice. First, in the PCB header, the power pins were lengthened and the sideband pins shortened so the power pins mate first and break last: the sense pins only engage once the power pins are seated. Second, the SENSE0 and SENSE1 definitions were rewritten.
Those sense pins are the part almost nobody checks, and they are the part that decides whether your GPU is told the truth. Four power levels are encoded on them, 600, 450, 300 and 150 W, and Intel requires both the PSU and the card to support them. Three details from Table 3-6 and section 3.3.1 are worth memorising:
- The 150 W level is signalled by a short between SENSE0 and SENSE1 presenting 0 Ω at the cable plug. Those pins must remain floating with respect to ground; if they are also shorted to ground, Intel notes the card “would misread the available power as 600 watts instead of 150 watts.”
- Open on both pins means no power may be drawn. Intel is explicit that in the Open-Open high-impedance state, no power may be drawn by the load.
- A modular PSU must configure the sense pins inside the PSU, not in the detachable cable, so that interchangeable double-ended 12V-2x6 cables work correctly.
That last point is the practical test for an adapter. If you are running a new GPU on an older PSU through a third-party adapter, the sense-pin wiring is being decided by a part the PSU manufacturer never specified. A unit with a native 12V-2x6 cable removes the question entirely, and that is the single strongest reason to buy a 3.1 unit rather than cable your way around one. The PSU tier list sorts units by build quality once the connector is settled.
Transients and hold-up: the two timing numbers that matter
Graphics cards do not draw a steady number of watts. They spike, and ATX 3.1 formalises how much they may spike. Table 3-1 sets the allowance at 3× the card’s maximum sustained power over a 100 µs interval, decaying logarithmically through the range until it reaches 1× at one second. A card rated at 450 W may therefore pull well over a kilowatt for a fraction of a millisecond, repeatedly.
This is why a PSU that passes a steady-state load test can still shut down in a game. The protection circuit sees a transient it was not designed for and trips. Buying for transient behaviour is not a marketing story; it is the reason the excursion table exists in the specification at all.
The second number is hold-up time, and it is the one I would put on every PSU box. Section 4.2.9 requires the PSU to maintain output regulation for 12 ms at 100% of full load, and recommends 17 ms at 80% of full load after a loss of input power. That is the window your machine has to ride through a gap.
If you run a UPS, this is the number that decides whether the UPS actually protects you. Mains drops, the PSU holds for 12 ms, and the UPS has to detect the loss and transfer to battery inside that window. A UPS with a slow transfer and a PSU that only just meets the minimum is a machine that reboots during every brownout. The UPS runtime calculator covers the battery side; the hold-up side is the PSU’s job, and it is worth preferring a unit that meets the recommended figure rather than the required one.
Efficiency: the spec floor is lower than you think
The gap between the ATX minimum and the sticker on the box is enormous, and it confuses a lot of buyers. The specification’s required efficiency is 70% at full load. An 80 PLUS Gold unit is far above that. Both statements are true, and neither one is a lie, because the ATX figure is a compliance floor and 80 PLUS is a voluntary certification program run by CLEAResult that tests at 20%, 50% and 100% load and grades the result (CLEAResult).
Cybenetics ETA is a second, independent certification with its own test methodology and its own grading (Cybenetics). Corsair’s own explainer is a fair summary of why the two schemes sometimes disagree on the same unit: they measure different things at different load points with different tolerances (Corsair).
What this means for a purchase is simple. Read the certification and the test report, not the ATX version, and not the wattage. Bel Fuse’s technical paper on the 80 PLUS tiers makes the same point from the enterprise side: the certification is a procurement shortcut, and the levels above Gold exist because the efficiency curve at low load is where the money actually goes (Bel Fuse).
For a desktop that idles most of the day, the low-load number is the one that shows up on your bill. The power cost calculator will do the arithmetic with your own electricity rate.
How to size a PSU for a 2026 GPU
Size from the connector budget and the transient allowance, then round up. Three rules cover almost every build.
Rule one: one native 12V-2x6 cable per high-power card. The connector is rated to 600 W sustained, and the sense pins tell the card what the PSU can actually deliver. Do not plan a build that needs two 600 W cables unless the PSU genuinely provides two, with the sense pins configured for each.
Rule two: leave headroom above the card’s rating, not above its average. ATX 3.1 permits 3× excursions over 100 µs. A PSU with a healthy margin and good transient response handles those without tripping; a unit run at 90% of capacity does not have anywhere to go.
Rule three: treat 675 W as a ceiling, not a target. Intel raised the maximum for a single add-in card from 300 W to 675 W. That does not mean your card draws 675 W, but it does mean the specification now contemplates cards that would have been unthinkable on a 3.0 unit.
The unit I would start from for a single-GPU build is Seasonic’s FOCUS GX-750. Seasonic states that the FOCUS GX series complies with the ATX 3.1 and PCIe 5.1 standards and covers 750 W to 1000 W, so the 750 W model sits at the bottom of a documented 3.1 range rather than at the top of a 3.0 one (Seasonic).
I may earn a commission if you buy through these affiliate links. As an Amazon Associate I earn from qualifying purchases.
Search AmazonLink checked 2026-11-16 Search NeweggLink checked 2026-11-16The PSU wattage calculator handles the sum for a specific parts list. Use it, then add the margin you are comfortable with rather than the margin a retailer suggests.
| Situation | What I would buy | Why |
|---|---|---|
| New build, one GPU with a 12V-2x6 input | 750 W class unit with a native 12V-2x6 cable | Native cable means the PSU configures the sense pins, per Intel’s modular-supply requirement |
| Existing 3.0 PSU, new GPU | A 3.1 unit, not an adapter | The adapter decides the sense-pin state, and a misread costs you either a shutdown or a 600 W assumption on a 150 W cable |
| Small case, short cable runs | Check cable length before wattage | A 12V-2x6 cable that will not reach is a 0 W cable |
| Machine on a UPS | Prefer the recommended 17 ms hold-up over the 12 ms minimum | The UPS transfer has to fit inside the hold-up window |
What I would check before buying
The order matters, because the first check eliminates most of the market.
- Does the unit have a native 12V-2x6 cable, and does the manufacturer document its sense-pin configuration? If the answer is “we include an adapter”, keep looking.
- What does the efficiency certification actually say? 80 PLUS level and, where available, a Cybenetics ETA rating. Ignore the ATX version here.
- What is the hold-up specification? If the manufacturer publishes 12 ms at full load and nothing else, it meets the minimum. If it publishes a figure at 80% load, compare it to the 17 ms recommendation.
- What is the warranty and who honours it? A ten-year warranty from a company that will still be answering email in ten years is worth more than a higher efficiency grade.
- Does the physical unit fit? Length, cable clearance behind the motherboard tray, and the number of 12 V EPS connectors your board needs.
If the machine is already built and behaving badly rather than being planned, the PSU failure symptoms guide covers the diagnostics, and the RTX 5080 PSU guide works through a single high-power card in detail. Neither one is a substitute for reading the label on the unit you already own, which is where most of these questions actually get answered.
Frequently asked questions
What is the difference between ATX 3.0 and ATX 3.1?
Is 12V-2x6 the same as 12VHPWR?
Do I need a native 12V-2x6 cable, or is an adapter fine?
What does hold-up time mean on a PSU?
Does an ATX 3.1 badge mean the PSU is efficient?
How many watts do I need for a modern GPU?
Can I use a 12VHPWR cable in a 12V-2x6 socket?
Why does my PC shut off during a power flicker even with a UPS?
Sources and corrections
- Last updated
- Methodology
- See our methodology for research and review standards. It draws on 8 cited sources, listed below, each checked against the original page on the date above. The specification figures here are read directly from Intel’s ATX Version 3.1 Multi Rail Desktop Platform Power Supply Design Guide, document 336521 revision 2.1a, which states in section 1.1 that it aligns to ATX Specification Version 3.1. Section and table numbers are given inline so you can check each number in the PDF. Efficiency certification descriptions come from CLEAResult, Cybenetics and Corsair’s own explainer, not from retailer copy. Seasonic’s FOCUS GX ATX 3.1 page is cited only for the existence and wattage range of that product family. I have not load-tested a power supply for this page. Every source was opened live on 2026-11-16.
- Sources
- Intel: ATX Version 3.1 Multi Rail Desktop Platform Power Supply Design Guide, doc 336521 rev 2.1a accessed 2026-11-16
- Intel Design Center: ATX version 3.0 multi-rail desktop platform power supply design guide (document landing page) accessed 2026-11-16
- PCI-SIG: Specifications (PCI Express CEM revision index) accessed 2026-11-16
- CLEAResult: 80 PLUS program details accessed 2026-11-16
- Cybenetics: ETA certification program accessed 2026-11-16
- Corsair: 80 PLUS vs Cybenetics, what the efficiency certifications mean accessed 2026-11-16
- Bel Fuse: 80 PLUS, why you need titanium power supplies in your data center accessed 2026-11-16
- Seasonic: FOCUS GX ATX 3.1 series product page accessed 2026-11-16
- Corrections
- Spotted an error? Email a correction with the affected line and what you found. The draft includes this page’s address. Verified changes are recorded here. Contact details.