Monday, August 29th 2022

ATX 3.0 PSU Specification Loophole Lets Manufacturers Evade Stringent Excursion Tolerance Testing by Simply Excluding the 12+4 pin Connector

When testing recent high-end GPU launches such as the AMD Radeon RX 6950 XT graphics card, we observed some system reboots that caused due to power-draw spikes (technically known as "excursions"). This is when a graphics card, for a brief moment, draws more power than the connector is capable of supplying, triggering the PSU's electrical protections, and causing a reboot. The new ATX 3.0 specification for the next-generation of desktop PC power supplies enable PSUs with a high tolerance for excursions, and prescribes testing standards to ensure a PSU meets the ATX 3.0 spec. Aris Mpitziopoulos (crmaris) and Jon Gerow (Jonnyguru) discovered wording in the latest Intel ATX 3.0 specification that could make you pay closer attention to reviews of ATX 3.0-spec PSU, when they're out.

In a Hardware Busters article, Mpitziopoulos clears the confusion that the ATX 3.0 power supply specs by Intel and the PCI-Express 5.0 power specification by the PCI-SIG, are mutually-inclusive—they're not. An ATX 3.0 PSU is not required to include a 12+4 pin (or 16-pin) ATX 12VHPWR connector, whether or not their nameplate Wattage is above 450 W, which means it's subjected to passing less stressful transient-response tests, particularly the ability to deal with a 200% excursion for at least 100 µs.
The ATX 3.0 also introduces a feature called Alternative Low Power Mode (ALPM), which is hardware-level preparation for OS-level system low-power device states, such as Windows 10 Modern Standby and ChromeOS Lucid Sleep. The ATX 3.0 spec had prescribed certain new hold-up time requirements for PSUs offering ALPM, however it was discovered that these specs only apply to the 3.3 V rail, and not the all-important 12 V.
Put simply, the vagueness in Intel's specs allow the manufacturer of a cheap/mainstream PSU to claim ATX 3.0 readiness by excluding a 12VHPWR connector at any nameplate-Wattage, just so they can evade the steep transient-response testing required to include such a connector. It allows them to [over]compensate with a large number of conventional 150-Watt 8-pin PCIe connectors, and hope that people use cables included with their next-generation graphics cards that convert a number of PCIe connectors into a high-Wattage 12+4 pin connector.
Sources: Hardware Busters, Igor's Lab
Add your own comment

23 Comments on ATX 3.0 PSU Specification Loophole Lets Manufacturers Evade Stringent Excursion Tolerance Testing by Simply Excluding the 12+4 pin Connector

#1
TheinsanegamerN
Oh no! like, all 3 people out there with a 12v board are going to be so inconvenienced :laugh: :roll: :laugh:

Meanwhile the rest of us wont have issues because we still use the 24 pin connector, like the rest of the AIB industry.
Posted on Reply
#2
btarunr
Editor & Senior Moderator
TheinsanegamerNOh no! like, all 3 people out there with a 12v board are going to be so inconvenienced :laugh: :roll: :laugh:

Meanwhile the rest of us wont have issues because we still use the 24 pin connector, like the rest of the AIB industry.
Future high-end Gen5 graphics cards will inflict excursions. If you use a PSU that doesn't meet those tolerances, your system will randomly reboot when gaming.
Posted on Reply
#3
TheinsanegamerN
btarunrFuture high-end Gen5 graphics cards will inflict excursions. If you use a PSU that doesn't meet those tolerances, your system will randomly reboot when gaming.
MMMmmm.... jus tlike nvidia's 12 pin connector that never saw use outside of nvidia founder cards?

Something tells me the industry is gonna stick with the usual 8 pin for awhile yet.
Posted on Reply
#4
LFaWolf
So if a PSU does not have the 12+4 pin, then don’t buy it, am I reading this right?
Posted on Reply
#5
btarunr
Editor & Senior Moderator
LFaWolfSo if a PSU does not have the 12+4 pin, then don’t buy it, am I reading this right?
If a PSU yells "ATX 3.0" on the tin and lacks 12+4 pin connector, pay attention to its reviews, or choose another PSU that has one.
Posted on Reply
#6
Nanochip
So in other words, when buying an "ATX3.0" power supply make sure to get a version with the the new 12+4 pin connector... and ignore all else.

Does anyone understand why certain standards setting bodies include vague requirements and/or loopholes? I mean look at the USB nightmare. USB3.0, USB 3.1, USB 3.2 Gen1/Gen2/Gen2x2. What the actual F. Why not just simple labeling: USB 3 (5, 10, 20 Gbps)?

And in regards to power supplies, do some of the member companies like Nvidia, PSU manufacturers etc., have too much sway into the language of the standards... so as to not negatively impact future sales?

I can't understand why some PSUs that don't have to undergo the stringent testing requirements are able to be considered "ATX3.0"?

What's the functional difference between existing ATX2.x PSUs (e.g., Corsair RMx1000) and an "ATX3.0" PSU (that lacks the 12+4 pin) connector?
Posted on Reply
#7
_Flare
An ATX 3.0 PSU is not required to include a 12+4 pin (or 16-pin) ATX 12VHPWR connector, whether or not their nameplate Wattage is above 450 W, which means it's subjected to passing less stressful transient-response tests, particularly the ability to deal with a 200% excursion for at least 100 ms. "Should be 100 µs" like in the shown table, right?
Posted on Reply
#8
Kenjiro
Even without 12VHPWR connector, PSU must work with 150% power excursion, if it wants to be certified to ATX 3.0 spec.
Some older PSUs did shutdown with even slightest overload.
Posted on Reply
#9
ThrashZone
Hi,
Talking about high end gpu's and referring to a lower powered 450w psu is pretty funny comparison to me frankly :kookoo:

Not to mention most spec's say 850w recommended.
Posted on Reply
#10
Daven
Woohoo! Race to the bottom!
Posted on Reply
#12
DeathtoGnomes
btarunrthe ATX 3.0 power supply specs by Intel and the PCI-Express 5.0 power specification by the PCI-SIG, are mutually-inclusive—they're not.
Its been shady right from the start it seems, kinda like Privacy EULAs.
Posted on Reply
#14
Oshadorin
The Asus ROG Thor II lineup, announced 10 months ago, is still barely available anywhere, and do not have a native 12+4 pin connector, only an adapter. Whereas the lower tier ROG Loki lineup, do have the native PCI-E 5.0 connector. There is definitely a problem without production of RTX 3.0 PSU.
Posted on Reply
#15
WorringlyIndifferent
TheinsanegamerNMMMmmm.... jus tlike nvidia's 12 pin connector that never saw use outside of nvidia founder cards?

Something tells me the industry is gonna stick with the usual 8 pin for awhile yet.
Nvidia's 12 pin connector was a one-off, brand exclusive that they tried for a single generation of GPUs, and not even all the GPUs of that gen. The 16 pin 12VHPWR is specified by PCI-SIG, an industry-wide consortium that sets (among other things) connector standards.

Scream and cry all you want, but the 16 pin is going to be the new standard, just like USB-C replacing USB-A.
Posted on Reply
#16
taka
Real soon we should see 1 Farad capacitors like the ones used in auto amps.
That is one cheap way to get rid of the spikes.

Posted on Reply
#17
zlobby
takaReal soon we should see 1 Farad capacitors like the ones used in auto amps.
That is one cheap way to get rid of the spikes.

Caps don't get rid of any spikes on their own.
Posted on Reply
#18
80251
Bypass capacitors can get rid of spikes (and dips I guess) but I thought the problem being addressed here wasn't so much spikes as dips in supply voltage.
Posted on Reply
#19
trsttte
TheinsanegamerNOh no! like, all 3 people out there with a 12v board are going to be so inconvenienced :laugh: :roll: :laugh:
ATX3.0 is not ATX12VO, 12VO is just a compliant possibility/option
Posted on Reply
#20
jonnyGURU
TheinsanegamerNOh no! like, all 3 people out there with a 12v board are going to be so inconvenienced :laugh: :roll: :laugh:

Meanwhile the rest of us wont have issues because we still use the 24 pin connector, like the rest of the AIB industry.
Wow. You've been here for 10 years and don't know the difference between ATX12VO and 12VHPWR?
TheinsanegamerNOh no! like, all 3 people out there with a 12v board are going to be so inconvenienced :laugh: :roll: :laugh:

Meanwhile the rest of us wont have issues because we still use the 24 pin connector, like the rest of the AIB industry.
"Something".
NanochipSo in other words, when buying an "ATX3.0" power supply make sure to get a version with the the new 12+4 pin connector... and ignore all else.

Does anyone understand why certain standards setting bodies include vague requirements and/or loopholes? I mean look at the USB nightmare. USB3.0, USB 3.1, USB 3.2 Gen1/Gen2/Gen2x2. What the actual F. Why not just simple labeling: USB 3 (5, 10, 20 Gbps)?

And in regards to power supplies, do some of the member companies like Nvidia, PSU manufacturers etc., have too much sway into the language of the standards... so as to not negatively impact future sales?

I can't understand why some PSUs that don't have to undergo the stringent testing requirements are able to be considered "ATX3.0"?

What's the functional difference between existing ATX2.x PSUs (e.g., Corsair RMx1000) and an "ATX3.0" PSU (that lacks the 12+4 pin) connector?
My attitude is: If Cybenetics or Intel (or whatever 3rd party that can test to the spec) doesn't say it's ATX 3.0 compliant or even PCIe 5.0 compliant, don't believe it.
takaReal soon we should see 1 Farad capacitors like the ones used in auto amps.
That is one cheap way to get rid of the spikes.

I thought the same thing... but then realized the inrush just to charge the cap would shut down the PSU before you even had a chance to POST! :D :D :D
Posted on Reply
#21
80251
How long does a 1F capacitor hold charge?
Posted on Reply
#22
jonnyGURU
80251How long does a 1F capacitor hold charge?
1F cap would be overkill. :D

A 1/4F would be enough for a 1500W PSU, but even that would trip the inrush current on charge. :D
Posted on Reply
#23
80251
Wouldn't it be possible to limit the inrush current to the capacitor by using a series resistor?
Posted on Reply
Add your own comment
Nov 21st, 2024 07:19 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts