The short version
- A PSU's wattage rating is its capacity. Your PC only draws what its components ask for.
- Efficiency varies with load, so where your system sits on the PSU's curve matters more than the label.
- Brief GPU power excursions are real, and Intel's ATX specification now tests for them.
- Age and heat wear PSUs down, so a margin on day one helps on day 1,000.
- Roughly 30% over your estimated draw is Khan's rule of thumb. It is a sensible starting point, not a standard.
Capacity versus consumption
Khan makes a useful point: a 1000W unit in a 350W system won't pull 1000W. The components set the draw and the PSU supplies it. The risk on the other side is a PSU with no margin. It works hard, runs hot and spins its fan up.
Khan's efficiency example uses 80 Plus Gold. The certification table for 115V internal, non-redundant supplies requires 87% efficiency at 20% load, 90% at 50% and 87% at 100% for Gold. I'm paraphrasing the table here, so check the exact figures against the certifier's listing for your model.
That makes the article's arithmetic correct:
- An 800W load at 87% efficiency needs about 920W from the wall.
- A 400W load at 90% efficiency needs about 444W from the wall.
The difference is lost as heat. That heat is why a PSU running near its limit tends to be louder and to warm the case.
Caveat: these are certification minimums, not a curve for every Gold unit. Real PSUs vary by model. Khan says efficiency is "fairly flat" from 20% to 80% load, and that exact claim doesn't hold for every unit. Other builder guides put the sweet spot at 30 to 70 percent load for ATX 3.1 units. Those are retailer sources, so treat them as rough guidance. The best evidence is the measured efficiency curve for the exact model you're considering.
GPU power spikes
The strongest part of the piece is the point about transient spikes. Intel's ATX 3.0 design guide defines how much brief overload a PSU must tolerate. For supplies over 450W with a 12V-2x6 connector, the guide lists these excursion limits as a percentage of rated output, each with a test duty cycle:
| Duration | Excursion limit (PSU over 450W, with 12V-2x6) |
|---|---|
| 100 µs | 200% |
| 1 ms | 180% |
| 10 ms | 160% |
| 100 ms | 120% |
Those figures are test requirements for the PSU, not a claim that every graphics card doubles its draw. Khan's example of a 350W card hitting 700W is illustrative. He gives no GPU model or measurement, so don't read it as a rule. Other sources agree on the principle. One explainer notes that these spikes often last under 100 microseconds yet can still stress a PSU or trip its protections. Another is more careful: a card marked at 450 watts does not normally consume 900 watts for long periods.
The practical takeaway is narrow. A PSU's transient response and protection behavior matter alongside its continuous rating. Headroom helps, but it doesn't guarantee that nothing will trip. Nor does extra wattage on its own make a PSU safer. Check that the unit is ATX 3.1 compliant and comes with a native 12V-2x6 cable if your GPU needs one.
Aging and second-hand units
Khan warns that electrolytic capacitors dry out over time, which lowers capacitance and raises ESR. Seasonic's engineering notes back the general point: PSU longevity depends primarily on component aging, namely capacitors and fans, and operating temperature and load affect service life. Seasonic also stresses that a PSU's MTBF is a statistical measure, not a guaranteed lifespan.
Two cautions apply:
- An 800W PSU does not become a 700W unit on a schedule. Nothing in the sources supports a fixed rate of decline.
- Khan's opening claim that a used PSU "might be the worst idea" goes further than the evidence. The safer statement is that a used unit's age, history, condition and warranty are hard to verify. Treat it as a risk, not a certainty.
Seasonic also notes that 80 Plus certification covers efficiency and power factor, not ripple or service life. Cybenetics ETA rates efficiency and LAMBDA rates noise. None of these badges is a full safety or quality verdict.
Choosing a wattage
Khan suggests running your parts through Cooler Master's PSU calculator, which asks for your components and returns a recommended supply. Treat the result as an estimate rather than a measurement of your finished PC. Then:
- Enter your actual CPU, GPU, motherboard, memory and storage.
- Check the GPU maker's own recommendation, especially for high-end cards.
- Add margin for overclocking and for the GPU you might buy later. Khan targets nearly 30% above the calculated need. He also keeps a couple of hundred watts spare so a future GPU upgrade doesn't force a new PSU.
- Compare specific models by connectors, ATX version, protections, efficiency curve and noise data, not just brand or badge.
- If you buy used, verify the exact model, its condition and warranty status. Model-specific test data is better than a wattage label.
Does oversizing cost you? Khan says it only hurts if the system idles below roughly 10–20% of the PSU's capacity. That specific cutoff is his claim and isn't established by a standard. Still, the direction is right: an extremely large PSU in a light-duty PC may sit at a less efficient part of its curve. One ATX 3.0 guide notes that idle power has fallen so far that PSU loss at low load is a real concern, which is why newer specs include low-load efficiency requirements.
Verdict
Khan's central advice holds up: size the PSU above your expected draw, and don't confuse rated wattage with real consumption. His specific percentages (30% headroom, a 20–80% flat zone, a 700W spike from a 350W card) are rules of thumb. Use them as starting points, then check the exact PSU's efficiency curve, ATX 3.1 compliance and connectors. For a mid-range build, a modest margin over the calculator's figure is plenty. For a flagship GPU, the transient spec and connector matter more than the headline wattage.