What good is a spreadsheet full of wattage numbers if the unit you pick browns out at 3 a.m. and takes two GPUs with it? That is the trap most rig builders fall into. They compare power supplies by the big number on the box, pick the cheapest one that clears it, and then learn the hard way that rated watts and usable watts are two very different things.
Comparing mining rig power supplies properly means looking past the sticker. You need to weigh efficiency at the wall, how the unit behaves under sudden load swings, whether the 12V rails are split in a way that trips protection circuits, how hot the room gets, and what happens if the thing dies in month fourteen. This guide walks through each of those in the order that actually matters, with the practical shortcuts and the mistakes that cost people real money.
Start With Real Load, Not the Wattage on the Box
The first step in any comparison is a number most people guess: how much power the rig actually pulls. Add up the board power of every GPU, not the vendor’s “recommended PSU” figure. Then add the CPU, motherboard, fans, storage and risers. Risers draw from the 12V rail too, and they are easy to forget.
A useful rule of thumb is to size the supply so the rig sits at roughly 50–70% of rated capacity during normal mining. That leaves headroom for transient spikes and keeps the unit in its most efficient band. Running a PSU at 95% load around the clock is how you turn a five-year warranty into a two-year lifespan.
Watch for the rated watts versus continuous watts problem. Many units are rated at 25–40 °C ambient, and derating applies above that. A garage in summer is not a 25 °C test bench. If the datasheet does not publish a temperature rating or derating curve, that silence is itself a data point in your comparison.
One more sanity check before you go further: count connectors against draw. A 6-pin PCIe connector is nominally rated around 75 W and an 8-pin around 150 W. If a GPU needs two 8-pins and the PSU only has the cables for one and a half cards, the wattage number is irrelevant.
Efficiency: Read the Report, Not the Badge

80 PLUS is a voluntary efficiency certification launched in 2004 and now administered by CLEAResult. Tiers run from plain 80 PLUS through Bronze, Silver, Gold, Platinum and Titanium. Titanium, added in 2011, requires 90% efficiency at 10% load, 94% at 50% and 96% at 100% for 230 V input.
Here is the part that matters for mining economics: efficiency is measured at the wall, so a 90% efficient unit draws about 11% more from the socket than the DC load it delivers. At 50% load, a Gold unit typically wastes roughly a tenth of its input as heat. Over a year of 24/7 operation, that difference between Gold and Platinum is real money, and it also shows up as heat you have to move out of the room.
Two cautions when comparing badges:
- Efficiency varies by input voltage. A 115 V rating and a 230 V rating are not the same test. Do not compare a European badge against a North American one without checking the certification report.
- The badge is a floor, not a guarantee. Certification reports list the actual test conditions. A unit that barely scrapes Gold at 230 V may behave very differently in a hot room at 60% load.
If you want a structured way to weigh spec sheets against each other, the same discipline that applies when you compare CRM tools by which features actually matter works here: decide which criteria drive the decision, score every candidate on those, and ignore the marketing extras that do not affect the outcome.
Transients, Rails and the Connectors That Fail
Modern GPUs do not draw a steady current. They spike. The ATX 3.0 specification, announced in 2022, introduced the 12VHPWR connector and requires PSUs to handle short power excursions well above rated wattage. ATX 3.1, from 2024, revised that connector to 12V-2×6. For a mining rig running last-generation cards this may not apply directly, but it tells you something useful: transient handling is now a formal part of the spec, not a nice-to-have.
Then there is the single-rail versus multi-rail question. Multi-rail units split the 12V output into separate rails, each with its own overcurrent protection. Put a riser and a GPU on the wrong rail and you can trip OCP even though the total wattage is fine. Single-rail units avoid that headache but lean entirely on the unit’s overall protection. Neither design is inherently better; what matters is knowing the rail layout of the specific model before you wire it.
And the risers themselves deserve their own line in your comparison. SATA-powered risers are widely reported as unsafe for high-draw GPUs, and that guidance comes from community and vendor consensus rather than a formal standard, so treat it as strong practical advice rather than a regulation. Prefer 6-pin or 8-pin PCIe feeds. A PSU that cannot supply enough PCIe connectors for your risers is disqualified no matter how good its efficiency looks.
Build a Comparison Table That Actually Decides Things
Once you have three or four candidates, put them side by side on the criteria that change the outcome. A table like this beats a page of spec-sheet screenshots:
| Criterion | Why it matters | What to check |
|---|---|---|
| Continuous watts at temperature | Rated watts may not hold in a hot room | Datasheet ambient rating and derating |
| Efficiency at 50% load | Drives electricity cost and heat | Certification report at your input voltage |
| Rail design | Prevents OCP trips on risers | Single vs multi-rail layout |
| PCIe connector count | Limits how many GPUs you can feed | 6-pin and 8-pin totals |
| Transient handling | Prevents shutdowns on load spikes | ATX 3.x compliance, review data |
| Fan and noise profile | Affects room cooling strategy | Independent noise certification |
| Warranty and RMA terms | Mining is hard on parts | Length, region, advance replacement |
That last row is underrated. Quality units in the mining-relevant range typically carry 5–10 year warranties, but terms vary by region and model. A ten-year warranty from a brand with no local service center is worth less than a five-year warranty you can actually claim.
Reading independent load testing is the single most practical comparison method available. Reviewers who test with electronic loads and oscilloscopes publish efficiency curves, ripple, transient response and fan-noise data. Cybenetics runs its own efficiency and noise certification with published reports, which is a useful second opinion alongside the 80 PLUS database. When you are weighing tools that all claim similar specs, the approach in this guide to comparing AI coding tools before you commit transfers neatly: trust measured behavior over claimed behavior, every time.
The Mistakes That Show Up Again and Again
A few errors appear so often in rig builds that they are worth naming directly.
Sizing to the GPU vendor’s recommendation. Those numbers assume a gaming PC with one card, not eight cards running flat out. They are almost always too generous per card and too small in total.
Ignoring the wall circuit. A 1,600 W rig on a 15 A household circuit at 120 V is asking for a breaker trip at best. The PSU comparison is only half the electrical story.
Mixing PSUs of different ages on one rig. It works, but when one unit fails, diagnosing which one is causing intermittent shutdowns becomes a guessing game. Keep a log.
Buying on badge alone. Two Gold units can differ by several percentage points of real efficiency and by a wide margin on transient behavior. The badge narrows the field; it does not pick the winner.
Skipping the hold-up and regulation specs. The ATX specification defines minimum hold-up time (typically at least 16 ms at full load) and voltage regulation tolerances of plus or minus 5% on 12V and 5V, 10% on 3.3V and 5% on -12V. Units that drift outside those tolerances cause instability that looks like a GPU fault.
Forgetting resale and repairability. A unit with a common fan size and standard cables is easier to keep alive. Sealed proprietary designs are disposable.
If spreadsheets and spec grids are your idea of a good evening, you will enjoy the same structured thinking in a pipeline management tools comparison across sales stacks — the criteria differ, but the method of scoring candidates against weighted needs is identical.
Practical Steps for a Fair Comparison
Here is a repeatable process that keeps the decision honest:
- Measure or calculate the rig’s real DC load, including risers, CPU and fans.
- Multiply by 1.4 to 1.6 to find a comfortable rated wattage target.
- Shortlist units with published efficiency reports at your input voltage.
- Check the rail layout against your riser and GPU wiring plan.
- Confirm PCIe connector counts exceed your needs by at least one cable.
- Read at least one independent load test for ripple, transients and noise.
- Verify warranty length and local RMA support.
- Compare total cost of ownership over three years, including electricity.
- Buy one unit first, run it hard for a week, then standardize.
Step nine is the one people skip. Standardizing on a single model makes spares, cable kits and troubleshooting vastly simpler once a rig grows past two or three units.
Conclusion
Comparing mining rig power supplies is not a wattage contest. It is a question of how a unit behaves at the load point you will actually run, in the temperature your room actually reaches, with the connector layout your risers actually need. Efficiency badges narrow the field, independent load testing picks the winner, and the rail design and connector count decide whether the build works at all. Get those four things right and the PSU becomes the least interesting part of the rig, which is exactly what you want from it.
Frequently Asked Questions
How much headroom should a mining rig PSU have?
Aim to run the rig at roughly 50–70% of the PSU’s rated output during normal mining. That keeps the unit in its efficient band, leaves room for transient spikes, and reduces heat stress. Sizing at 95% load around the clock shortens lifespan even on quality units.
Is a higher 80 PLUS rating always worth the extra cost?
Not automatically. Going from Gold to Platinum can save several percentage points of efficiency, which adds up over 24/7 operation, but the payback depends on your electricity price and how long you keep the unit. Check the certification report at your input voltage before paying a premium.
What is the difference between single-rail and multi-rail power supplies?
Multi-rail units split the 12V output into separate channels, each with its own overcurrent protection. That can trip OCP if a GPU and riser land on the same rail. Single-rail units deliver 12V from one channel and rely on the unit’s overall protection. Neither is universally better; know the layout before wiring.
Are SATA-powered risers safe for mining GPUs?
They are widely reported as unsafe for high-draw cards, and the guidance comes from community and vendor consensus rather than a formal standard. Prefer 6-pin or 8-pin PCIe feeds for risers, and make sure the PSU has enough PCIe connectors before you buy it.
Why does my PSU shut down when the GPUs ramp up?
Sudden load spikes can exceed what the unit handles, especially on older designs without ATX 3.x transient headroom. Other causes include multi-rail OCP tripping, overheating in a hot room, or a unit rated at a lower ambient temperature than your setup reaches. Check the datasheet and the rail layout first.
Do I need to match PSU wattage to the number of GPUs?
No. Match it to measured total draw, including risers, CPU and fans. Connector count is the harder limit: a 6-pin PCIe connector is nominally around 75 W and an 8-pin around 150 W, so count cables against cards before trusting any wattage figure.