Hardware Insights Forum

http://www.hardwareinsights.com/eval ... thodology/

Discuss the article here

My thoughts:

• Given that you FAIL any unit that doesn't comply with ATX specifications, I think you can also FAIL any unit that doesn't meet even the basic safety standards (e.g. by using non-safety-rated capacitors on the mains).
• On the topic of safety, I don't like the cheap glass fuses that can only safely interrupt 10 times their current rating. (The mains is capable of delivering quite a bit more than 50 to 150 amps for that fraction of a second…) Personally, I think that high rupturing capacity fuses (which have a ceramic body and sand filling, and can interrupt up to 1.5kA) are the only type that should be used there.
• Short-circuit protection testing? Penalties for 2-transistor standby supply? What about if the bleeder for the X-caps is missing, misplaced (e.g. if there's an X-cap before the main switch but the bleeder is after it), or of too high a value to be effective?
• Of the saving technologies, I consider the thermistor bypass to be much more important than the X-cap discharge IC:

The bleed resistor (if properly chosen) will only waste a fraction of a watt. Through a few quick simulations, I found out that you need about a 430kΩ resistor to discharge 1µF worth of X capacitors in time (from 375V to 50V in 1 second), accounting for usual tolerances (capacitors 10% high and the resistor 5% high). Which gives us:

• For 0.47µF, a 910kΩ resistor (typical dissipation of 58mW at 230V and nominal value, worst-case dissipation of 81mW at 265V and value 5% down)
• For 0.68µF, a 620kΩ resistor (typical dissipation of 85mW at 230V and nominal value, worst-case dissipation of 119mW at 265V and value 5% down)
• For 1.0µF, a 430kΩ resistor (typical dissipation of 123mW at 230V and nominal value, worst-case dissipation of 172mW at 265V and value 5% down), as already mentioned
• For 1.5µF, a 270kΩ resistor (typical dissipation of 196mW at 230V and nominal value, worst-case dissipation of 274mW at 265V and value 5% down)
• For 2.2µF, a 180kΩ resistor (typical dissipation of 294mW at 230V and nominal value, worst-case dissipation of 411mW at 265V and value 5% down)
• For 3.3µF, a 130kΩ resistor (typical dissipation of 407mW at 230V and nominal value, worst-case dissipation of 569mW at 265V and value 5% down)

To size the resistor for other capacitance values, divide 430000 by the µF and round down to the nearest available value; for ±20% tolerance caps, use the next E24 resistance down. (I know the CAPZero datasheet says you can use much higher values, even with ±20% tolerance caps. But I'm assuming that's with the cutoff between hazardous and SELV levels at 120VDC, which isn't totally safe. My values are conservatively chosen.)

The important thing about the NTC thermistor is its value. 2.5Ω (±25%) or thereabouts is typical, and while adequate to prevent the PSU from self-destructing, you still can't say "low" inrush… Additionally, if the unit is running near maximum load, and the mains is interrupted briefly (then reconnected before the thermistor has time to cool down), the inrush can be even higher. The bypass circuit enables a higher resistance to be used, without the power waste (aside from the power used by the relay coil). This matters, if you don't like tripping circuit breakers when connecting the power to multiple units at once.

And given how many power cords go to waste, the least manufacturers can do these days is to enable us to opt out of getting a new one with each unit.

- besides that I almost do not review the cheapest crap anymore as nobody sends that to me, I think none of such will pass loading tests in spec and have non-safety caps; so, such crap won't pass anyway, and if so, I will deal with it when it happens
- that's matter of opinion, so far I have not yet seen any exploded, it always blew well inside the glass tube; and you still should have several circuit breakers on the way
- yeah I am testing it usually even during loading, it is somewhat tricky to measure the ripple and sometimes I short something; if not, I deliberatelly short it after 100 % load; 2transistor SB is nothing bad, it can be made properly and with the same efficiency as using PWM IC, I have seen such later FSP units; as for the resistors, I usually check for that
- thermistor bypass is more important, on the other hand, it is also problematic thing; it makes the NTC itself to survive almost indefinitelly, on the other hand, relays itself can burn pretty badly, so in a whole picture, you cannot say it is clearly positive as the X cap IC is

A glass fuse did explode in one of the el-cheapo units c_hegge reviewed. The whole idea of using a fuse rated for less fault current than that available just seems iffy to me.

It's not that 2-transistor circuits can't be properly designed. But they are primitive and most (all?) lack SCP, so short the standby output and the switching transistor will blow. IIRC, everell posted (on Badcaps.net) about a unit (FSP?) that used a Zener diode to clamp the output voltage, which of course shorted from excess power dissipation, and serious damage resulted to the PCB. They made reasonable sense back in the '70s and '80s (not that PCs had any standby supplies back then), but suitable IC switchers aren't exactly a new development.

Have you actually seen the relays fail in those high-end PSUs? As far as I know, it's rare for relays to fail when operated within their limits.

I've seen many bad relays in UPSes.

I have the datasheet for Panasonic's JW series relays which are a fairly typical model available in ratings of 5A (single-pole or double-pole) or 10A (single-pole only) with a resistive load at 250VAC. They are specified to last for 100,000 switching cycles at full load, and if de-rated can last up to 500,000 cycles below about 2A/3A, respectively (observing the graphs provided).

However, when removing the drive, the relay coil can produce a spike of several hundred volts easily (potentially >1kV!) in the opposite polarity from the applied voltage, as its inductance tries to maintain the current flow. This is bad as it will damage the controlling element and potentially even the (very thin) insulation of the coil winding wire.

Connecting a diode in parallel with the relay such that it is reverse biased by the drive voltage will clamp the spike to something negligible (1V or less), but will also slow down the release. This is not so good for the relay contacts. The release time can be improved by adding a resistor in series with the diode; the best performance is obtained by using a Zener diode in combination with the standard diode (connected back-to-back such that the Zener acts on the reverse voltage spike). See here for a useful guide.

EDIT: On second thoughts, that's not the main problem.

„with a resistive load at 250VAC“

That's it, well, amongst other things. The input capacitor makes it very far away from resistive load which still means spikes of 30-60 A (just check I think TPU is it who measures inrush current?) even with thermistor to supress. There is quite often audible spark. And if something gets bad and the relay gets to open/close escesivelly, it dies quite fast.

Besides, how many times have you seen Panasonic or Omron relays anywhere? I see only chinese crap these days.

That probably explains the failures you've seen, although I can't comment on build quality of relays.

Bypassing the thermistor should be easier on the relay, though, as there will only be a few volts across the thermistor once the primary capacitor(s) have charged.

Out of curiosity, how many failed switches do you see? I've read something here about mini DPST switches (at least the 10A+ types) having marginal spacing between the poles, so if the switch is abused, they can short together, either blowing the fuse or, if the switch is before the fuse (as is not uncommon, and is also allowed for X and Y capacitors), tripping/blowing the house circuit breaker/fuse (or blowing the plug fuse if BS 1363). My guess is that the wider double-pole types aren't prone to that problem.

See for example:

• 8800 is a narrow SPST type (also available in a dual unit, 0.5mm wider than a standard-width type)
• 8500 is a DPST type, like those in question
• 8600 (single-throw), 8610 (double-throw), and 8620 (centre-off) are standard-width single-pole miniature types
• 8650 (single-throw), 8660 (double-throw), and 8670 (centre-off) are wider double-pole miniature types
• 1250 is a large, but narrow, DPST type (which could have the same problem)
• 1500 (single-throw), 1510 (double-throw), and 1520 (centre-off) are the large (16A rated) single-pole type (also available in a dual unit, the same width as the double-pole types – 1550, 1560, and 1570 respectively)

Mind you, 15A (as rated by UL/CSA) does seem like a bold claim for the miniature switches. (They are the same agencies who chose to uprate the IEC 60320 connectors, though…)

Not really many bad switches, aprox. the same number as bad thermistor. It's like 1 of 100 bad PSUs or so. Once it was switch which was not turning on so the owner jumpered it to stay ON and once it was shorted internally thus stayed ON as well.

As for high-power UPS, quite often there are bad relays, but I think it is often because of other part failure which makes them cycle excessivelly.

Today is a great day as the long awaited and even longer planned AC breaker arrived. Now I will be able to break the AC input for any time I choose with 0.1 ms precision and measure the PSU hold-up time!

It is also time to move on with the evaluation methodology. I have already thought about it, I will keep the three market segment division, I think ti is appropriate. But I think it would be somewhat better to move the other price evaluation away and swap it for hold-up time. So like this:

-component quality (0–15)
-built quality (0–15)
-voltage regulation under combined load (0–10) and crossload (0–5, rated half)
-ripple under combined load (0–10) and crossload (0–5, rated half)
-efficiency (0–15)
-hold-up time (0–15)
-others, e. g. some special functions, accessories, longer warranty etc. (0–10)

Next to that I will make a performance-to-price ratio. With that everybody can calculate himself updated number years after the review came out when the price is completelly different. And probably some page with updated summary table so everybody can find it on one place and compare with others.

That is first question. The second question is what to do with units which fail the ATX minimum hold-up time. From TPU measurements, it seems there are shitloads of them. I am somewhat afraid that if I won't let such units pass, half of even high-end market would get FAIL award. So maybe generously setting some minimum under which the unit will fail but above such treshold still loose points hard when it is under ATX minimum?