Computer Build Resource Thread - Page 1351

80 plus tests power supplies for efficiency levels at certain loads: 20%, 50%, and 100%. As everyone's pointed out, efficiency varies by load. It also varies by the percentage of the load on +12V vs. +5V vs. +3.3V (and +5VSB). And it also depends on temperature and generally drops with higher temperature; 80 plus tests at 23C. If for whateve reason you're having the power supply intake air from inside the case, it could well be over 23C. The MOSFETs used for switching and maybe some other parts would have higher effective resistance at higher temperatures, just like your CPU and GPU. Thus you get more losses at higher temperatures, generally. Higher temperatures also reduce key components' ability to conduct current, so that reduces the max power that can be supplied. Sometimes they give different ratings at different intake temperatures, like 500W @ 40C, 400W @ 50C—though generally I don't think that's common anymore for consumer computer power supplies.

It's actually more efficient for a power supply to spend energy running a fan than it is to have the components running hotter. Hence why people are impressed by Seasonic Platinum (fanless) 520W's crazy efficiency results.

---
Now let's play the "Myrmidon does a lot of guessing and doesn't guarantee any accuracy" game.

The basic principle behind a linear power supply is that if the load isn't requiring much power, it will burn all sorts of excess power (i.e. ohmic losses in resistors) to keep the output around what it should be. Because this is a huge waste and also to keep transformer sizes reasonable, computers use switched-mode power supplies, aka SMPS. (Transformer needs to be larger if operating at lower frequency like 50/60 Hz as compared to something much higher, and also for higher output power levels. A huge transformer could easily cost more than entire high-end computer power supplies.)

The operating principle behind a computer SMPS is to (1) rectify input mains voltage into close to DC, (2) rapidly connect and disconnect (i.e. switch; and connecting/disconnecting is done by manipulating transistor drain/source/gate voltages, not mechanically) that rectified voltage to the main transformer so you get a fast square wave, (3) make a square wave of smaller magnitude appear on the other side of a transformer because that's what transformers do with AC signals, and (4) rectify / filter / convert the output of the transformer to +12V / +5V / +3.3V in some way or another and many other details.

The important thing to note is that there's a controller that determines how long per cycle the transformer is connected on the mains side. Whenever it's connected, energy is being transferred to the other side. If more power is consumed by the computer, more energy needs to be sent across. Also, all this switching action can't be done perfectly, so there is power lost there, though there are many hax improvements and ways to mitigate that power loss.

At low loads, not much energy needs to be transferred, and the "on" time per cycle is relatively low. A relatively high amount of power is lost needing to do all the switching. (That said, some power supplies have all sorts of hax tricks regarding how to operate the transformer, save energy. Some actually reduce the switching frequency at low loads, which I think can save energy at the expense of performance. This is why in some reviews you see the ripple is relatively high at lower loads, or maybe the ripple is lower at 40% than at 20% or something weird like that.)

Also, consider all the capacitors, maybe some other parts. The filter capacitors at say the +12V output are there no matter what the load is, and they'll always have about +12V. I think losses there are mostly due to the equivalent series resistance (low but nonzero), so there's leakage current losses there no matter if you're at full power supply load or a low load. The transistors and other ICs have some quiescent current draw even when off or not doing anything.

In other words, different sources of power losses scale differently with output power. At too low loads, some inevitable losses that don't really increase much (or less than linearly) with load, will be a relatively large power draw compared to output power delivered. Hence low efficiency. At high loads, you get higher internal operating temperatures and higher currents running through the circuit. Both increase the power draw of the switching transistors relative to power delivered, for example. So worse efficiency past a certain point.

Urgh, long post. Probably not even correct in a couple areas or missing a few obvious things.


edit: Efficiency is more related to the overall design, but there can be some difference at lower loads if you change from a higher-wattage version of one to a lower-wattage version (which should have same switching frequencies and many other common elements). The lower-wattage version may have some less ICs and their quiescent draws, maybe some fewer capacitors or lower values so less losses there, and so on. However, it's probably not really all that different between say a 450W model @ 100W output compared to the 650W model of the same series and design @ 100W output.

edit2: Hopefully somebody reads at least part of the above, lol.

Also, you can maybe get some ideas about tradeoffs between high efficiency and high performance. Use more filtering caps, lose a slight amount of energy there, and get lower efficiency but lower ripple. Lower the switching frequency and you get less losses at the expense of responding to changes in the output and regulating those at a slower pace (I think). LLC resonant designs minimize switching losses, but this is supposedly at the expense of quick transient responses. At least without some more voodoo black magic solutions that the top-tier PSU manufacturers know about, which enables high efficiency with all kinds of high performance (with high cost).