PSU Tier List 2026: Power Supply Tiers, Every Row Sourced

By LK Wood IV · Published 2026-07-29 · ~14 min read · St. Louis County, MO

Search for a PSU tier list right now and the top results include a Linus Tech Tips forum thread whose own title says [OLD], posted in July 2016; the Cultists Network list, whose current revision 17.0g is dated November 2023; and a scattering of YouTube Shorts. Power supplies moved on: ATX 3.1, the revised 12V-2x6 connector, and a generation of refreshed platforms arrived after most of those lists stopped being maintained.

This page is our answer: a tier database of 28 current power supply families, rebuilt from scratch for the ATX 3.x era. It has two rules the old lists never enforced, and one column none of them carry.

The rules: every specification traces to a source linked in its row — the manufacturer’s own page where one is reachable, otherwise the professional review that published the specification — and every tier placement leans on a test linked right next to it, meaning Cybenetics certification data and hands-on reviews from outlets that load-test on real equipment. Where we could not verify something, the cell is blank. Blank beats guessed.

The column nobody else has: what these units do at the low loads a 24/7 homelab actually runs at, where the 80 PLUS badge on the box is silent. More on that below the table.

How the tiers are decided

Tier letters are our editorial call, but they are derived from stated criteria, so you can audit any placement:

• Tier S — a top-of-class professional verdict, a 10-year-or-longer warranty, and either elite measured performance or elite efficiency and noise results. These are the units reviewers compare everything else against.
• Tier A — excellent units with strong professional verdicts and warranties of at least seven years, held out of S by one real finding: price, a transient-response result, a shorter warranty than the class leaders, or weak light-load efficiency.
• Tier B — good units with a caveat you should know before buying: a louder fan profile, an older ATX 3.0 revision still on shelves, a warranty we could not verify, or a rival that simply tests better at the same price.
• Tier C — the budget class. Fine choices when right-sized and bought at the right price, with bigger compromises: Bronze certification, fixed cabling, shorter warranties, or weak efficiency at the extremes.
• Avoid — reserved for units whose failures a professional outlet has documented and stood behind. Two entries qualify, both cited.

Four housekeeping notes, because honesty about method is the whole point of this page. First, several 2023–2024 lines were silently refreshed from ATX 3.0/12VHPWR to ATX 3.1/12V-2x6; where the professional review covers the older build, the row says which revision the verdict describes. Second, we name a unit’s OEM platform only when the linked review states it — PSU platform attribution is a folklore-rich area and we do not repeat folklore. Third, rows for units with no independent test behind them were left out; the ones we considered and dropped are listed at the bottom of this page.

The fourth note is an exception we want on the record rather than buried. One row — the EVGA SuperNOVA 1000 GT — carries no independent lab test, because no major outlet ever published one for the GT line. It is in the table anyway, in Tier C, because we ran one for an extended period in a working build and published our own long-term review. That is first-party experience, not a lab bench, and its row says so. It is the only row that rests on our own bench rather than an outlet’s, and if that distinction matters more to you than the hands-on account, treat the row as informational.

The 2026 PSU tier list

Filter by the wattage class you are shopping and the price band you can stomach. Every row stays in the page for search engines and AI assistants regardless of filters; the chips only hide rows on your screen.

How to read it

• Tiers rank quality, not size. A Tier S 750W unit is not “better” for your build than a Tier B 1000W unit if your build needs 1000W. Size first — our PSU wattage calculator does the arithmetic from your actual parts, and our sizing guides for the RTX 5070, RTX 5080, and RTX 5090 walk the GPU-specific cases. Then use this table to pick the best unit at that wattage.
• Standard and connector matter more than they used to. ATX 3.1 with a native 12V-2x6 is the current target for a modern GPU build. 12VHPWR-era units still work — the connector revision improved contact reliability margins, not compatibility. Watch the cable ratings printed in the rows, because the marking and the wiring do not always agree: Hardware Busters found MSI’s MAG A850GL shipping a 16-pin cable marked 600W whose sense pins were configured for 450W, and found the opposite on Thermaltake’s GF3 1000W, where the documented 450W figure sat against a cable the reviewer measured as configured for 600W. Trust a measured review over a printed number.
• Warranty is a quality signal. The industry’s best platforms carry 10–12 years because their makers expect them to survive it. A blank warranty cell means none of that unit’s sources stated one — which is itself worth knowing.
• Efficiency badges are a floor, not a curve. The certification column describes a unit at its tested load points, which start at 20% for every 80 PLUS level below Titanium. It tells you nothing about the loads an idle machine spends its life at — which is the next section.

If a unit you own is misbehaving — random restarts, black screens under load, coil whine that changed character — the tier list will not diagnose it; our PSU failure symptoms guide walks the actual triage.

The 24/7 homelab dimension nobody’s tier list covers

Here is the gap in every PSU badge on every box. The 80 PLUS program publishes minimum efficiency thresholds at 20%, 50% and 100% of rated load for every level; only Titanium publishes a minimum at 10% load, per Bel Fuse’s published threshold tables. Below the tested points the certification says nothing about how a unit behaves. Intel’s ATX v3 design guide addresses this directly, adding a low-load point at 10W (or 2% of capacity on units above 500W) that requires 60% efficiency and recommends 70% — the guide’s stated reasoning being that desktop idle power has fallen so far since 2005 that power-supply loss now dominates what an idle machine pulls from the wall.

Now put that next to what always-on hardware actually draws. ServeTheHome measured a Minisforum MS-02 Ultra at 19W idle at the wall and a ZimaCube 2 Pro at 25–30W; a typical N100 mini PC runs single digits to low teens. A homelab box spends the overwhelming majority of its hours at exactly the loads the badge never measured.

The measured numbers make the point better than the argument does. Cybenetics publishes full light-load tables for the units it certifies, and down at a genuine idle load, current Gold-and-above 750–1000W units span 68% to 78%:

All figures at 115V, from each unit’s Cybenetics evaluation report (every report linked in Sources). The load column states the exact point measured: Cybenetics’ low-load row is 2% of rated capacity on units above 500W — 20W on a 1000W unit, 17W on an 850W one — and its light-load table also publishes a fixed 20W point. This list covers the units with published Cybenetics light-load data, which is not the same set as the tier table above; the FSP Hydro G Pro is here for the measurement, not as a tiered recommendation.

Two practical conclusions fall out of this table.

Right-size an always-on box even harder than a gaming PC. Fifty watts on a 1000W unit is a 5% load — below every point the badge tested, and near the bottom of every curve above. The same 50W on a smaller quality unit sits meaningfully higher up its curve. Four families in the table reach down to 550W — the Pure Power 12 M, Century II Gold, Smart BM3 and CX750 — and our wattage calculator will tell you whether your parts list fits one. Resist the instinct to buy double.

When the box runs 24/7, pick from the top of the low-load column. The spread above is real but modest — at a 30W idle, the gap between the best and worst rows works out to a handful of watts at the wall. What makes it worth acting on is that an always-on node multiplies those watts by 8,760 hours a year, across every node you run, and the same units that measure well down here tend to be the quieter, better-built ones anyway. Our 24/7 homelab idle power costs breakdown has the measured draws and annual math for the hardware itself; the power cost calculator turns your own wattage and electricity rate into a yearly figure. And if the box matters enough to run 24/7, it matters enough to put behind a properly sized UPS.

One honest boundary: for a gaming desktop that idles a few hours a day, none of this moves your bill meaningfully; a few watts across a few idle hours is noise. This section exists because homelab duty cycles are the inverse of gaming duty cycles: 90-plus percent idle, every day, forever.

Your PSU is not in this table. Here is how to tier it yourself

The database covers 28 current families, so plenty of perfectly good older units are absent. The method behind every row above is reproducible in about five minutes, using the same public sources:

1. Find it in the Cybenetics database. Search cybenetics.com for the exact model. A per-unit report gives you measured average efficiency, the light-load table this page leans on, noise in dBA, and standby draw — far more than the badge on the box. No listing at all is itself information: it means no independent lab published full-range numbers.
2. Read what the platform is, from a review rather than a forum. A hands-on review from an outlet that load-tests will name the OEM platform and how it behaved at temperature. If the only thing you can find is a spec sheet and opinions, treat the unit as untested rather than good.
3. Check the warranty against the class. Current quality platforms run 10–12 years. A 3–5 year warranty on a modern unit is the manufacturer telling you what it expects.
4. Check the connector against your card. Confirm the 16-pin cable’s actual rating, not its marking — two units in this table have a mismatch between the two.

If it clears all four, it belongs somewhere in Tier A or B whether or not we have listed it. If it fails step one and step two, that is the real warning sign — not a missing tier letter.

The questions builders keep asking

Should I be worried about my PSU’s tier?

If your current unit is running a system it is sized for and behaving normally, no. Tier lists earn their keep at purchase time, when the units in a price bracket look identical on the box. An installed, stable, correctly sized unit from a reputable maker does not become risky because a spreadsheet ranks something else higher.

Is a lower-tier PSU dangerous with a high-end GPU?

Tier C in this table means compromised, not dangerous. Every C row rests on a test — an outlet’s, or in the single flagged case above, ours — and where a protection is missing, the row says so outright: the MSI MAG A750BN’s entry records that LTT Labs found no under-voltage protection, which is the kind of detail you want printed rather than smoothed over. What deserves genuine caution is the Avoid class, where the failures are documented rather than theoretical, and mismatched sizing: a 300W-rated 16-pin cable feeding a 350W card is a real problem no tier letter fixes. Check the connector column, then check your wattage math.

Should I replace a working eight-year-old power supply?

Not on age alone, but the calculus shifts once a unit is well past its warranty. Capacitors degrade with heat and hours, and an old supply that was sized for the card you had in 2018 is often genuinely undersized for a modern one — which is the condition that stresses it hardest. If it is out of warranty, feeding a much bigger GPU than it was bought for, or showing any of the failure symptoms, replace it proactively rather than after it takes something with it. If it is comfortably sized and behaving, leave it alone.

What is deliberately not here

• Seasonic Vertex PX, Super Flower Leadex VII Platinum PRO — manufacturer specs verified, but we found no independent lab test of those specific lines. Spec-sheet-only rows do not get tier letters here.
• Seasonic Prime TX and be quiet! Dark Power Pro 13 — these now exist only at 1300W and above, a halo class this table does not cover. Both are well-regarded; neither is a mainstream buy.
• Montech Century G5 — gone from Montech’s own catalog with no professional review we could find; its successor Century II Gold is in the table instead.
• Whole brands, and this is scope rather than judgment — Thermalright, Lian Li, ASUS, Gigabyte’s current lines, Antec, SilverStone and Phanteks are absent because v1.0 research covered the vendors above, not because their units were assessed and rejected. They are first in line for the next revision.
• Successor lines we saw but did not verify — be quiet! Dark Power 14 and Pure Power 13 M, MSI’s GS-suffix refreshes beyond the A750GLS, FSP’s ATX 3.1 Hydro refreshes. They join the table when a verifiable test exists; that is what a living database means.
• OEM folklore — platform attributions you may have seen elsewhere for units whose linked reviews do not state one.

Revision 1.0 — published 2026-07-29, 28 units. Next scheduled review: October 2026. Corrections and additions are welcome at hello@techfuelhq.com — bring a source and we will check it against the same standard we held ourselves to. The underlying dataset is downloadable as open JSON under CC BY 4.0, and it carries the same revision stamp, so you can diff it against any later version.

Sources

Certification and specification primaries: Intel ATX Version 3 Multi Rail PSU Design Guide, doc 336521 · CLEAResult 80 PLUS program details · Bel Fuse 80 PLUS threshold tables · Cybenetics ETA methodology · Corsair 80 PLUS vs Cybenetics explainer

Low-load measurements: Cybenetics evaluation reports for the Corsair RM850x, Corsair RM850e, Corsair SF750, Seasonic Focus GX-750, MSI MAG A850GL, be quiet! Pure Power 12 M, be quiet! Dark Power 13, FSP Hydro G Pro, FSP Hydro Ti Pro, XPG Core Reactor II, and XPG Core Reactor II VE

Homelab idle draw: ServeTheHome — Minisforum MS-02 Ultra power consumption · ServeTheHome — ZimaCube 2 Pro power consumption · HomeLab Starter power consumption guide · XDA on PSU oversizing and the efficiency curve

Per-unit review and specification sources are linked directly in each table row.