It's unfortunate how tenacious this 'common wisdom' is.  Maybe it was translated from home wiring circuits (where it's def important to maintain 20% headroom when talking about sustained loads,) or it's from older/cheaper/lower-end PSUs w/ more bell-shaped curves, but current high end PSUs have no issue running at capacity.  In fact the graph you pasted shows measurements up to 12.5% above cap - though I wouldn't necessarily test that w/ a 24/7 load.  

As to the efficiency and power use concern, peak eff is around 33% of cap - that would be insane to leave so much unused cap.  Assuming 75-80% is the guideline as suggested, you're talking about maybe a 2% eff drop between that level and full cap.  Which assuming a $0.1/kwh power rate translates into:

((1600 * .02) / 1000) * 0.1 * 24 * 365 = $28

That's $28 per year in extra energy cost by fully utilizing your PSU vs handicapping it at 80%.  Even if you double that (splitting the load will get closer to peak on each PSU,) given that that the AX1600i runs about $470 new, and the avg lifespan of electronics is 3-5 yrs, it's clearly much more financially prudent to simply utilize the full potential of the PSU.

Also, if someone is really worried about a reset-situation kicking the GPUs back to stock, they should just use PPTs.  Although even w/o that, the over-current protection of the PSU will handle what should be a rare occurrence anyway.


I found 1575 cclock was roughly the line for 935 mclock in my case.  Though this was a while ago, on an early bios and older drivers.