OEM information’s pretty widely available, including in Cybenetics reports.
Seasonic switched to direct sales years ago because they were bad enough their OEM contracts dried up. But somehow they still have good rep with a lot of forum posters.
It’s my understanding ASRock acquired EVGA’s PSU division. Some of the fan control’s wonky but the Phantom Gaming and Taichi have remote 12+4 OTP, which is perhaps the single best mitigation available short of Nvidia actually doing something about 12+4. Big jump from RM750e to TC-1300T, though.
be quiet!'s supplies are pretty mixed, so are best taken on a model by model basis. My experience is there’s usually a better option, though occasionally Dark and Straight Power get tempting pricing here. Also the rail and Y cable arrangement’s unusual, so a bit of extra planning’s wise if going be quiet!.
As @Susanna’s brought up, the updated header’s only half of a 12+4 connection.
While PCIe 5.1 does not change the 12VHWPR plug dimensions, cables are relabeled as 12V-2x6 because people commonly assume 12VHWPR cables need replacing too. The relabeling’s actually a PCIe spec violation but it’s not like PCI-SIG’s been showing a lot ability to think things through here. The header change is the sense pin retraction, so you can tell by looking.
Since we’re on the topic, also ATX 3.1 is a downgrade from ATX 3.0. So if a supply passes ATX 3.0 it passes 3.1. But since people assume newer version better ATX 3.0 supplies have got their specs and boxes changed to say 3.1.
12+4 headers are present on many PSUs. Since people assume newer is better a corollary is PSUs without a 12+4 connector are perceived as bad. So supplies with 12+4 have a way of selling better even though 2x8 → 12+4 is better engineering.
I don’t know of any PSU released with a 12+4 header within the 12VHPWR timeframe, so probably they’re basically all 12V-2x6, but if you look hard enough you might be able to find one that’s not a prototype or engineering sample.
Measurements of GPU-cable-PSU assemblies show Nvidia’s 600 W claim is broken by design. Doesn’t help PCIe 5.0/5.1’s both internally inconsistent and incompletely margined. But 12V-2x6 provides Nvidia a pretext to pretend 4080, 4090, 5070 Ti, 5080, and 5090 aren’t safety margin regressions from 8-pin power.
For 12V-2x6 to have the same safety margin as 8-pin and EPS it’d be rated at 275-300 W, depending on hardware details. However, 12+4’s smaller size means that’s likely optimistic due to higher power density and reliance on ground return primarily through the slot. So data presently available suggests more for ~250 W. It’s possible to construct reasonable arguments down to ~200 W or so. IMO 200’s pretty conservative on current data but the more 12+4 gets looked at the worse it gets. So ¯\_(ツ)_/¯
.
A 5070 pulls ~250 W through 12+4. If dGPUs weren’t so hard to get I’ve suggested returning the Prime for a 9070 upthread.