On 7/1/11, Robert Heller <heller at deepsoft.com> wrote: > At Fri, 01 Jul 2011 16:25:33 +0200 CentOS mailing list <centos at centos.org> > wrote: > >> >> Colin Coles wrote: >> perhaps naively, I'm surprised: doesn't this mean they put crappy PSUs >> in those servers? >> I thought decent PSUs were expected to deal with dirty input AC? > > AND *I* thought *switching power supplies* (effectively) rectified the > AC input and then used the DC to drive a higher frequency system to get > the desired output voltages. (The higher frequency means smaller, more > efficient transformers and need smaller filter caps -- all of which > means a lower cost, more reliable, more efficient power supply.) Which > suggests that both the input voltage and frequency are not particularly > critical, so long as it does not have massive spikes/surges or > consistently low voltage. In a normal SMPS that would be true, because the typical ATX PSU normally has two bulk input capacitors. However the better PSUs nowadays and those in servers are usually active PFC units which only has one. Now the problem occurs because non-true sinewave UPS usually use PWM to achieve the output. This means the UPS outputs a consistent high voltage but switches it on/off to achieve the same average power, i.e. 400V for x msec, then 0V for x msec = 200V average, where x should be much smaller than 16 msec IIRC, which is the required hold up time for ATX specifications. In a cheap and arguably badly designed UPS, the selected voltage is much higher than the PSU is expected to ever handle from a true sine-wave source (which is nominally 320V peak for a 230V RMS source). So it either blows the input capacitors (typically 400V values), or protective circuitry shuts it down first. On the non-PFC PSU, the voltage is divided across the two main caps so this isn't a problem.