Mr Evil wrote: If it can't reach the specified MTBF at full load continuously then the manufacturer is lying because the ATX spec says that's what the PSU must be able to do. Maybe no PSUs do live that long under full load, but in that case there are no true ATX PSUs available.
As long as the maximum power is not exceeded, it doesn't matter how much power is drawn; only the temperature matters.
Where in any ATX spec guide is there a MTBF requirement? I cannot find it, despite perusing the
Power Supply Design Guide for Desktop Platform Form Factors, Revision 1.1, March 2007, produced by Intel.
The ONLY mention of reliability in that guide, which is the Bible for designing ATX power supplies, is this (found on page 47 of the guide):
9 Reliability
9.1 Reliability - RECOMMENDED
The de-rating process promotes quality and high reliability. All electronic components should be designed with conservative device de-ratings for use in commercial and industrial environments.
Electrolytic capacitor and fan lifetime and reliability should be considered in the design as well.
Notice the big RECOMMENDED in this section? When Intel puts recommended in their guide, it's not a requirement for getting the designation as a qualifying ATX power supply, but is instead "the status given to items within this design guide, which are not required to meet design guide, however, are required by many system applications. May be a required item in a future design guide."
And all this MTBF, MTBF, MTBF....like each and every power supply or other component that lists a MTBF is supposed to last that long.
MTBFs are usually a bit of a con. They don't go around testing hundreds of units over a long periods of time since they can't afford the costs and don't have time to spare.
Basically, what happens is - they know or determine the reliability index for the weakest point (or points) in the product and approximate the reliability projections for the product itself based on that.
Let's suppose that the "normal" usage lifespan for a component is judged to be three years, and the MTBF is quoted as 1,000,000 hours. What that means is that if you take 1,000 of the components and run them all for three years, you would expect roughly 26 of them to fail within that three year period.
How do I work that out? If you run 1,000 of them for three years, the total number of component-hours is 3 * 365 * 24 * 1,000 (roughly 26 million). The MTBF is 1,000,000 component-hours in normal usage, i.e. in devices within the first three years of their life, hence about 26 of your 1000 components would be expected to fail during the three-year trial.
What an MTBF of 1,000,000 hours most emphatically
does not mean is that if you just take just one of the components and run it continuously it would typically run for 1,000,000 hours before failing.
The way the number is calculated depends heavily on what the manufacturer judges to be the "normal" component lifespan. In the above example, if they decided that people typically only used the component for one year, then an MTBF of 1,000,000 hours would mean that roughly 9 out of every 1000 components would be expected to fail in the first year. But, crucially, it could well be that a hell of a lot more than 9 would be expected to fail in the second year.