Power Supply Voltage Tolerances and Measurements

[Janitor Tzap]

Hee hee.....

The great bad capacitor plague.

I have a friend who works for a electronics parts supplier.
Had warned me about the bad capacitors coming out of China a few years ago.
Unfortunately there are still a lot of these defective capacitors floating around.
{Especially in the surplus market.}
So, check the date code on the caps.
Anything from late 2008 to the present should be ok.
Also go with a good brand name; Panasonic, nichicon, or United Chemi-con.


Signed: Janitor Tzap

[MrAl]

Hi again,


Ron:
I am keeping an eye open for nylon without pins.
That's interesting about the bad cap plague, maybe these caps came from that era.
As i was saying, i never saw caps get that bad so fast, but that would explain it.
The caps were not that hard to remove really...just four screws to take the lid off
of the PS itself, then four more to get the PC board out, then unsolder. The hugh
wire bundle is a bit of a pain, and seeing the values of the caps is hard until i
got them out because they are facing in a direction that prevents seeing their
values because there is something else in the way like another cap or a heat sink.
The space is minimal on the board itself on the component side, but on the copper
side it's not bad because it's just a flat plane, so unsoldering wasnt hard at all.
Took about five minutes or less to remove three caps. Will take less to solder
new ones in place


Janitor:
Yes hopefully the new caps will be better quality. The article said that there was
basically no way to test them for that bad type either, too bad. I hope i get
something decent this time. Makes you wonder.


Generally:
Oh yeah i checked the board layout a tiny bit and yes those chips do seem to be
the common TL494 or similar and the LM339 or similar. These two chips are very
well documented on the web in data sheets. I think i actually have some of the
TL494 chips laying around too from years back.
This means i could probably trace out the whole circuit board and do a schematic,
although it would not be that easy because there are lots and lots of parts. Im
not sure yet if i will do this or not.


ADDED LATER:
Parts ordered. Estimated repair date will be 10/10/2009.

[MrAl]

Hello again,


In a word, "Fixed!".

Ok, that doesnt do this little project justice either

The parts came in today (much sooner than expected) so i pulled the other two caps and
soldered in the 5 new caps. Two were 1000uf 10v and two were 2200uf 10v and one was 2200uf 16v.

I tested the supply and this time the +12v line supports lots of current, and the power supply does not
turn off. Before the repair almost any load on the +12v line would cause the power supply to
immediately turn off. Although i have not yet tested it fully, i believe this tells me that it is now working
properly. Of course more tests would be nice too so my next question has to do with that...

When testing the various outputs, what ratio does the +3.3v and +5v lines have to the +12v line
with respect to current? In other words, it would not be good to load the +3.3v line with 10 ohms
(which is only around 300ma) while loading the +12v line with 12 amps for example, because that is
not how the power supply operates in the typical computer. Any ideas how these outputs should
be loaded for a 'proper' test?


BTW, here are some tests that were performed on the capacitors in question. The waveforms
tell the story as the bad cap has a huge amount of ripple while the good cap has very little.
Dont forget to compare the vertical scales in each figure as they are very different. The
bad cap has something like 100 times more ripple than the good cap.
The huge peak to peak ripple across the bad cap shows that it's capacitance is way too low now,
and the straight vertical part of the waveform (the vertical straight part is not curved but perfectly
straight up and down) shows the ESR of the cap, and we can see that the ESR of the bad cap
is much greater than the ESR of the good cap because the vertical part in the left waveform is
much longer than the vertical part in the right waveform, which is the resistive part showing up.

[reloadron]

Way to go MrAl!

Also runs with your well thought out theory that the reason the PSU would not start outside the case even with a tiny load but would start inside the case was because of the additional capacitance offered by the motherboard.

As to loading or actually a cross loading ratio. I can tell you what I generally did to test the things. I worked from the name plate data. I looked at the maximum current ratings for each rail (voltage) and worked from there. However, keep in mind That if you look at the current ratings for each rail and multiply times the voltage then add all the watts you will get a number much larger than the unit is capable of. Though it should give you a reasonably good cross loading ratio to work with for cross loading as a test.

Ron

[MrAl]

Hi again Ron,


So i guess what you are saying is that if the nameplate reads "10A, 10A, 10A" (all voltage supplies
put out an equal max current) then we would load with the same current for each supply such as:
3.3v,1A
5.0v,1A
12v,1A
or
3.3v,5A
5.0v,5A
12v,5A

but if the nameplate reads "10A,10A,20A" for the 3.3v,5v, and 12v respectively, then we would load:
3.3v, 1A
5.0v, 1A
12v, 2A
or say
3.3v,5A
5.0v,5A
12v,10A

to keep the ratios the same, right?

This would mean the load would always have to have three parts, one for each + supply, and
the ratios would have to be planned out.

I can see it is going to be a bit difficult to test this supply thoroughly outside the case with no
motherboard as various very low value resistors are going to be required. This thing has specs:
+12v, 16A
+5v, 35A,
+3.3v, 24A
which comes out to 446 watts, and the power supply is marked as 450 watts, but this means
i would need:
12/16=0.75 ohms on the 12v supply and
5/35=0.143 ohms on the 5v supply and
3.3/24=0.138 ohms on the 3.3v supply.
I'll have to see if i have resistors that can handle this, but i dont think so. I could probably get
the 12v supply loaded to 1 ohm and that would be 12 amps, close but not quite 16 amps.
I'll have to check my rheostats to see if they go that low.

BTW, any one ever hear of a "Gear Max" power supply?
That's this one.

LATER:
Well it turns out that i dont have the proper resistors to test this anyway.

I almost forgot to mention something very important for anyone else going to fix their power
supply...
The caps that i ordered were the same height as the caps to be replaced, but the 2200uf
caps all had wider cases. The cases of the old caps were 10mm diameter and the cases of
the new caps were 12mm diameter. I figured i could rig these to fit in the power supply
i have, but for other power supplies this may not work and the correct size may have to be
found. I couldnt find 10mm diameter case caps so i went with the 12mm case size.
I had to install one of the 5 caps with longer than usual leads for a pc board, by using
heat shrink tubing on one lead that was about 3/8 inch long. This means when the cap
is installed it sits 3/8 inch above the board rather than sitting right on top of the board.
This means the ESR is going to be very slightly higher and the ESL is also going to be
a tiny bit higher, but i figured the bandwidth of the whole thing is lower than any of
those small increases would affect anyway (switching frequency of these supplies is quite
low really) and the motherboard caps would take care of the high frequency needs of the
motherboard. With the leads 'extended' on that one cap alone the other two caps that
mounted very close to that one could mount very comfortably.
The parts on these boards are mounted very close together so i was lucky that with this
supply the 12mm diameter caps worked ok, but as i said i wanted to mention this because
with some supplies the part packing density may be even tighter meaning the 12mm caps
may not fit in the case. The leads can be extended a little, but then you have to have
enough overhead height in the case to allow this. In some cases the leads can be bent
before installation too, which would allow the cap to lie down on the board rather than
sit upright, but that would require enough room to do that too.
It's also a good idea to mark the positive terminal with a red dot on the very top of the
case so that AFTER installation the polarity can be double checked before soldering.
It's very hard to see the polarity marking after the cap is installed with all the other parts
in the way, so it helps to be able to quickly check the polarity from the top. If one of
those caps gets installed backwards by accident it would be a disaster, so great care
is needed there to insure proper polarity.

[MrAl]

Hi again,

I wrote the above and then thought that it may be helpful to present a better graphic for the testing
of the power supply caps and since the above was getting a bit long in itself i thought it would be better
to post another message rather than extend that message again.

Here is a better picture of what the capacitance tests tell us...





One more little thing...

The total cost of the repair would have been about 5 dollars not counting shipping, but i ended up buying
a few more things while ordering the caps and some caps where cheaper when buying 10, so it came up
to a bit more than that, but the total parts actually used were 90 cents for two of the caps, 75 cents for
another, and 1.20 ea for two others, so that's a grand total of $4.95 USD. Not too bad i think.

[reloadron]

Something to note as to loading the unit. I would gradually add loading to the PSU. Meaning if the PSU 12 Volt rail was rated at 20 amps I would gradually increment up to 20 amps (or close to it) in steps. For example I used 2 Ohm 55 Watt resistors in series (4 Ohms) and incremented in 3 amp steps. The old drawing of the prototype still exist and can be found here. I also still have some images of the original prototype and those can be found here. Yes, to do this one needs some low resistance high wattage power resistors. When the actual unit was done I had steps for adding load to each voltage rail and also I had a step that was variable using some pretty beefy powerstats.

The PSU you have is actually pretty low current compared to many today and also interesting is note the emphasis of current on the 3.3 volt rail.

Additionally when I am at work as in now, I can't see images in post. I hate that part. LOL

Ron

[MrAl]

Hi Ron,


Wow, nice test set there! When i was pricing the resistors to test this kind of power supply properly
i didnt like the prices i was seeing ha ha. Maybe i'll look around again.

My rheostats are still a little high in resistance, the lowest being 6 ohms. It will handle considerable
wattage but it's wire wound and to get 0.2 ohms it would be turned down to it's last little wire loop.
Im sure that little loop can not handle a huge power for very long and maybe not even long enough
to make a quick measurement.

Where do you buy your big power resistors?

[haklesup]

How about some 12V lamps. Halogen or automotive. Still cost a few bucks a piece but then you can use them as replacements in your car or desk lamp. You can think in Watts or convert back to Amps.

Rheostats make great voltage dividers but lousy current sources. You will definately burn out the last few turns if you turn the dial that low. (Been there, Done that)

[reloadron]

HI All

MrAl

My rheostats are still a little high in resistance, the lowest being 6 ohms. It will handle considerable
wattage but it's wire wound and to get 0.2 ohms it would be turned down to it's last little wire loop.
Im sure that little loop can not handle a huge power for very long and maybe not even long enough
to make a quick measurement.

Where do you buy your big power resistors?

Years ago with the same company I am with now I was in the Navy Ordinance division. I worked with the MK46, MK48 (Including the MK48 ADCAP Advanced Capability) and the MK50 Torpedo programs. We developed and built the alternators used on those torpedo programs. Every alternator was tested extensively (which is humorous considering an alternator in a torpedo is going to actually have a short life span in real use). When the programs shut down as there are enough totpedos out there and the remainder of the MK50 went to Westinghouse I snatched the alternator test stands from scrap with all the spare parts for literally a few bucks. Thus I had a large collection of low resistance high wattage resistors which I figured someday would come in handy. Those Ohmite StackOhm resistors in the images are expensive. When I built the first load banks after the proto in the links I used Vishay aluminum housed power resistors 50 Watt in the 2 and 5 Ohm ranges. Those were purchased from Allied Electronics and ran about $3.00 each. The larger load bank was funded by the guys who wanted to test PSUs.

Here is what I did on the variable legs of the load bank. I always placed a fixed resistor in series with the rheostats. That served to limit the current on any leg of the load bank. For example if I place a 2 Ohm 100 watt potentiometer in series with a 2 Ohm fixed 50 watt resistor the maximum current the pot will ever see is 3 amps. Well below the 8 amps it is rated at even when we get down to those few final turns. We don't want to see our pot begin to glow cherry red. I hate that part. I also had fans for cooling. All that power we dissipate is going somewhere.

Haklesup brings up another good suggestion for loading a PSU and that being the use of 12 volt incandescent lamps. Problem being you need a heck of a lot of lamps and if you go high intensity halogen you need ceramic sockets. Additionally wipe the glass envelop down with some alcohol to be sure it is squeaky clean of any finger oils. They do actually provide a good load but when we get outside of 12 Volts like the 5.0 and 3.3 volt rails they don't fare too well. Something I have used everyday 120 VAC incandescent bulbs for is testing UPS (uninterrupted Power Supply's). Using 25, 60, 75 and 100 watt bulbs worked great for that. You would be surprised that a 100 Watt bulb actually does draw 100 watts with an uncertainty of less than 1 watt. I was surprised at that.

Ron

[MrAl]

Hello again,


Hackle:
You know that light bulbs change resistance quite a bit right? Quite a long time ago
i checked several types of bulbs and noted about a 10:1 change in resistance as they
heated up. A 100 watt light bulb for example has cold resistance around 10 ohms
and hot resistance around 100 ohms. This means that while the bulb is heating up
it is drawing a lot more current then we might like. Maybe only use them for the
5 and 3.3v supply lines?
Come to think of it though i dont remember what the typical automobile 12v bulb
draws now anyway.
I guess it could work though if the power supply can stand it...have you ever tried
this with a computer power supply?

Ron:
Oh yeah, 100 watt light bulbs do make loads for inverters as long as the inverter has
a good current limit circuit and it is able to start up into a higher than normal load.
If not, it might get stuck in current limit.


Also thought about using a hefty transistor as load, but varying the base current with
another transistor and a pot. Problem is i would need a big heat sink anyway.
Nice thing is it would be programmable to just about any current level.
I'll have to look around and see what i have in this high rating.
Oh yeah, i did find some power resistors for about $5 each on Mouser, but do you think
it is possible to load the supply at start up with maybe 75 percent of full load current
on each of the three main lines (5,3.3, and 12v)? I ask this because then i can
get away with maybe $15 worth of resistors.

Another thing that comes to mind which is very relevant here is just how much abuse
can a wire wound power resistor take? I noticed that although i used 2 ohms to test
the +12v line for the 'after repair' tests, the total rating was 20 watts (two 1 ohm
10 watt resistors in series) but they didnt heat up nearly as fast as i thought they would
and i was able to complete the test, taking less than a minute to perform.
The total current was 6 amps, so the total watts dissipation was of course 72 watts,
and that was the heating power getting to the 20 watt resistor yet it took some time
for it to heat up (didnt time it though, but perhaps i should repeat that test).
This is important because if i can get away with 1/4 wattage ratings doing a quick test
then that might be good enough.
On the other hand, that means i wont be able to do an overnight burn in test which is
a typical test for quality power supplies...run overnight at full load. I would need the
correct resistors for this or something to take the place like hackle suggests.

[haklesup]

I thought more about the bulb after I posted and yes Mr Al, a bulb would present a much lower resistance load initially. I suppose that might be a bit harsh when testing a suspect, weak or recently repaired supply but it would serve perfectly well for a overnight burn in with inherent visual feedback about the stability of the current.

I think one could mitigate the inrush current when the bulb warms up with a series inductor. It still wouldn't be linear but at least one could take the edge off the turn on current spike. I'm not sure its worth the effort though. you would only need one coil near the supply not one for each bulb. This also assumes a suspect supply didn't have significant ripple or the inductor would interact. apparently a resistor is the most ideal test load but a bulb cold be substituted under certan circumstances.

I've heard that a higher speed processor can go from very high supply current to very little is just a few 10s of clock cycles. A motherboard is far from a static load.

Sorry I don't have a good source for cheap resistors aside from the usual electronics junk shops. I salvaged most of mine from pre digital era equipment over many years. I don't have one handy but I wonder what is the resistance of a common electric range element, or a hairdryr heater etc. I suspect a just a bit too high

[reloadron]

Haklesup:

I've heard that a higher speed processor can go from very high supply current to very little is just a few 10s of clock cycles. A motherboard is far from a static load.

When a system first is powered up there is considerable current draw on the motherboard and other hardware like drives including optical drives and hard drive(s). Once things are stable and the system is not really doing much of anything the overall power draw gets somewhat stable. This is when CPU usage drops to and hovers around 2% to 3%. THis leads to exactly what haklesup mentions. The CPU will use a stepping process to up clock or down clock the processor based on demand and the Vreg for the processor will supply power accordingly. The end result is that the power draw on the PSU is literally all over the place.

I have a little device called a Kill A Watt which if Googled will bring up countless hits. Actually a pretty good device for measuring line side power to a PSU or other device. OK I have a system sitting here with 6 hard drives and two optical drives, the processor is a relatively recent Intel Q9450 quad core CPU.

With the system off but plugged in and using 5 Volt SB (Stand By) power the power draw at the AC mains is about 2 Watts. Press the power button and the system immediately jumps to a draw of about 220 Watts for about 3 to 5 seconds. This decreases down to about 165 Watts once we get through the BIOS and into Windows. Eventually dropping to about 140 Watts once the entire boot is complete and the system is basically sitting here doing nothing with a 2% CPU usage.

When I begin running a folding application and use 25% of my CPU the system clocks up and power draw is about 180 Watts. Adding another client for 50% CPU usage gets us up to about 200 Watts or an increase of about 20 Watts. If I add yet another client and run the CPU up to 75% I only see an increase to about 210 Watts. Unfortunately my 4th client to get me up to 100% CPU usage is not cooperating this morning so I can't really tax the CPU at 100% and see what happens. Bummer on that note.

Now we keep in mind that we are looking at line side so the actual PSU is not delivering that much power to the overall system as the PSU does have inefficiency which is likely the heat pouring out the back of it. Matter of fact I did some test measuring a system inlet air temperature versus the outlet temperature as well as PSU inlet and outlet temperatures. Remember that on most systems a PSU draws its cooling air from inside the case. My basic observations on this were that the PSU was drawing in air about 20 degrees F. higher than room ambient. This would be all well and fine but opens a new can-O-worms in that many PSU manufacturers spec their rated outputs on a psu running in a 23 Deg. C. ambient or about 73.4 Deg. F. Therefore the PSU is derated before it does much of anything.

On a bright note today's processors use considerably less power to do more work than their counterparts of a few years ago. This results in less heat being generated by the CPU. Intel has dozens of white papers on that subject listing the power demands of their CPUs.

Matter of fact one of the biggest consumers of power in an extreme system is the GPU (Graphics Card or Graphics Processing Unit). These GPUs frequently run in pairs and have additional 12 Volt connectors on them for additional 12 Volt power. Systems like this need power supplies capable of delivering considerable 12 Volt power. However, the average home system has little need for a PSU rated at over 500 Watts.

The next animal to consider is a PSU has plastered on the box that it is a let's say 500 Watt PSU. That really means little if the bulk of the power is dedicated to the 3.3 and 5 Volt rails. Today's systems rely more on 12 Volt power and that is where we want to see current capability. If we peek into a PSU we generally will see two transformers. There is a large one and a small one. The little guy is for the 5 Volt SB power rail and is active anytime the PSU is plugged in and the rear power switch is turned on. The remaining larger transformer serves all the output voltage rails using a series of taps for the voltages. This brings us to another problem Looking at MrAl's PSU:

12 Volts @ 16 Amp
5 Volts @ 35 Amp
3.3 Volts @ 24 Amp

You might get close to 16 Amps on the 12 Volt rail with no to minimal load on the remaining rails but I will guarantee that you will never be able to load all 3 rails above to their maximum and not have the PSU literally fold over. A good PSU will fold over, a bad or real cheap PSU will likely just smoke and die.

MrAl if you want some load resistors to mess with you are welcome to borrow them including some potentiometers. I will muddle through some of what I have lying around here and let you know what I have. I'll send them to you and when you are done experimenting just send them back this way.

Ron

[MrAl]

Hi again,


Ron, that's very generous of you. I'm still looking around a bit so let me see if i cant find something so
you dont have to go through all that trouble. I've derated a bit after reading your replies, so here is
what i am going to go for i think:

3.3v, 0.2 ohms, which is 16.5 amps which is 55 watts
5v, 0.2 ohms, which is 25 amps which is 125 watts
12v, 1 ohms, which is 12 amps which is 144 watts

Now that's only a total of 324 watts, so i would think a power supply marked
with "450 watts" should be able to do it, right?


BTW, that's approximately what my system does at and after turn on Ron.
I had measured the input power on my system too and found that yes the power
demand goes up and down a lot depending mostly on processor usage.
The disk drive usage didnt affect the power that much.

[CeaSaR]

Ron,

Where did you get your Kill-A-Watt? My local "The Shack" has them on the shelf (ok, rack).

CeaSaR