Posted On: Mon, Mar-05-2007
Written By: Oklahoma Wolf
Title: Capacitors and the Computer PSU
Introduction: Capacitor. The very word conjures up profound indifference amongst my family members every time I use it in conversation, which is surprisingly often. But, as many of you are aware, they are of profound importance to the operation of a Switch-Mode Power Supply, or SMPS for short. Today we will be taking a close look at how different quality capacitors can affect a computer power supply's output.
Capacitor. The very word conjures up profound indifference amongst my family members every time I use it in conversation, which is surprisingly often. But, as many of you are aware, they are of profound importance to the operation of a Switch-Mode Power Supply, or SMPS for short. Today we will be taking a close look at how different quality capacitors can affect a computer power supply's output.
An SMPS is inherently a very noisy device. Not noisy as in listening to my local fire siren from ten feet away mind you (I'll never do that again), but noisy as in alternating current showing up on the DC outputs where we don't want it. It is impossible to eliminate all such noise due to the very nature of SMPS design, but through good filtering we can reduce it to an acceptable level so our motherboards and hard drives don't cook while we're fragging some poor shmoe in the latest hot shooter or playing Oblivion for the forty billionth time through.
I will be experimenting on two patients in this article. The first victim, er... PSU is an Antec Neopower 480W constructed deep in the factories of Channel Well Technology. The second unit is a first generation Fortron-Source FSP600-80GLC Epsilon I originally purchased shortly after its Canadian debut and then promptly traded for a Silverstone ST56ZF. Both of these units, in their own way, have been the cause of much discussion and speculation among the PSU sections of the more well known computer hardware forums due to some interesting test results and high profile failures - more on that as the article unwinds.
My experiment on these units will take the form of what initially seems a very simple procedure: take out the old capacitors and put in better ones. But, there are a few things to keep in mind when doing so. First, the quality of the replacements is important - we want high grade, low ESR (Equivalent Series Resistance) capacitors with nice long lifespans. Not just any brand will do. And we want them to be rated at 105 degrees Centigrade, for you see 'tis the nature of a computer power supply to run at warm and sometimes very warm temperatures.
For this article, I chose KZE and KY model capacitors from United Chemi-Con, a well respected company that has been in the cap business for decades. Some other excellent capacitor manufacturers are Matsushita (Panasonic), Rubycon, Nichicon, Samxon, and Sanyo. These can be obtained in several places online around the world; my own source is usually www.digikey.com where they have an excellent selection.
Before we go any farther, in order to replace capacitors you must be handy with a soldering iron. While soldering and desoldering capacitors is not hard, it does require some basic familiarity with soldering. Yours truly has been doing it regularly since age fourteen, so what is second nature to me may prove difficult for you. If you are uncertain you can do this successfully, please don't try until you've practiced on something you can afford to destroy; like the next door neighbour's constantly cranked stereo or something. Just don't get caught, and make sure you have an alibi.
And now, some tools of the trade:
Most of these are pretty easy to identify. We have a spool of solder, small needle-nose pliers, a screwdriver, a massive Soldapullt solder sucker, a 40 watt Archer soldering iron, and a terminal block with a needle in it. You might be asking, "A what with a what in it?"
This is my preferred desoldering tool for computer mainboards and other stubborn multi-layer PCBs. Because this is a stainless steel sewing needle, solder doesn't stick to it. This makes it handy for use around surface mount devices where that big blue solder sucker is perfectly capable of wiping out whole rows of tiny components with a single slip. It also offers the advantage of not dropping little solder boogers all over what your working on if you've forgotten to clean it.
For this article, I didn't use the needle in the terminal block due to being very experienced with the big solder sucker, and also for the reason that neither PSU had a multi-layer PCB to complicate things. I also didn't use that 40 watt iron above, opting instead for a more powerful 60 watt borrowed from a friend. I'll explain why later on.
A quick word about the soldering environment: while a computer PSU usually will not contain many static sensitive components, if any at all, it is always a good idea to solder using a wrist strap, grounded soldering iron, anti-static mat, or a combination of the three. I like to live dangerously, so I went with none of the above. Do as I say, not as I do!
Here we have a picture of the enemy, three failed capacitors that can cause all sorts of fun things in the computer while you're trying to get your game on, from spontaneous restarts to strange hissing noises to failing to power on to giving you a light show to rival the best Independance/Canada Day fireworks. The middle capacitor is actually from an Epox mainboard, but I tossed it in there anyway so Fuhjyyu didn't think I was exclusively picking on them.
Now, without any further ado, let's move on to our first victim.
Here we have our first patient, an Antec Neopower 480W owned by the good CAD4466HK from our very own forums. These CWT built Antec units have come under fire lately due to a higher than average rate of reported failures. The problem has gotten so bad, some speculate, that Antec has been forced to go to a new source for their units (Seasonic). Why is there such a big problem with these? Simply put, these are built for silent operation with fans that run accordingly slow. Capacitors hate getting hot. While they are passive components, they generate their own heat and if not kept cool, they can fail early. CWT has traditionally been very fond of using Fuhjyyu capacitors, which can perform well, but are more sensitive to heat than most.
Over the past few years I have owned four power supplies from the CWT factory, only one of which is an Antec. Only one has never needed new capacitors, and it is an ancient unit lacking thermal fan control. All used Fuhjyyu and Teapo for capacitors. It may surprise you to learn that the longest lived of the three that did need replacement caps was actually the one Antec branded unit - it went three years before needing the operation. This Neopower is only a few months old, which is really far too young to showcase this issue, but it will be interesting to see if replacing the capacitors inside will affect its performance all the same. Teapo is widely thought of to be quite decent for use in computer power supplies, so we won't be too hard on them. Start heating your iron now... we want it good and hot for the next couple hours or so.
I am now sealed stick will be lost or removed. It shall be out of warranty validity. It goes without saying that getting anywhere near the caps will void the warranty, but forewarned is forearmed. Or maybe it's the other way around... my forearms don't seem to think so. Ha! I hear groaning... maybe I should forewarn you about forthcoming jokes... naaaaah.
With the cover off, we can see the primary side (input) on the left and secondary (output) side on the right. It is the secondary side where we'll be spending our time, as replacing the big primary filter cap will more than likely do absolutely nothing for performance. These rarely fail, and if they do you have bigger problems than dead caps.
Mmm, pasghetti. Somewhere in this mess are several Fuhjyyu capacitors we're going to replace. But to do it, we have to pull the PCB completely out of the housing.
In preparation for the pulling of the PCB, it's a good idea to do some labeling if working on a modular unit like the Neo. Then, carefully remove the connectors from the metal.
Hey! Big yellow EMI filtering cap on the AC input! You're in the way, dude! I can't get the PCB out of there until you move!
Much better, thank you. See those black and white wires heading in the direction of the PCB? Those are looking to keep us from removing the PCB thanks to their short length and will have to be desoldered. But, before doing so, grab yourself a sheet of paper and a pen and make one of these:
A diagram and list of all major capacitors you plan to replace inside the unit, and note where all wires go that you plan to desolder in order to pull the PCB. By "major capacitors," I mean all capacitors with a value greater than 400uF. Make sure you note the polarity on your diagram, as it is possible to run into the odd silkscreen on the PCB that was printed backwards at the factory. On this PSU however, this isn't a problem. Installing a capacitor backwards and then powering up the device will immediately produce a loud bang as the cap instantly blows up. Trust me, when it happens it's not fun. Man, do my ears hurt... kidding, just kidding, put down the axes.
A better look at the wires needing to be desoldered. In order to do this, unscrew the PCB and carefully lift it out from the opposite end. Set the guts down in such a way that you can get at the solder pads shown in the center of the below pic, heat 'em up, and pull the wires out. Then, use your preferred desoldering method to clean up the holes for easier resoldering later.
Free of the enclosure, we can now readily gain access to the underside of the PCB. If you're concerned about getting zapped, you can now safely discharge what's left in the primary side capacitor through the use of a 2 kilohm 2 watt resistor across its solder pads. Or, just don't touch that area, which is the approach I take. Usually, these big caps will be discharged by now, but better safe than sorry. Like I said, I live dangerously so I didn't bother discharging. Do as I say, not as... wait, I said that already didn't I?
While we're looking at this picture we can see why I dumped the 40 watt iron for the more powerful one. The traces attached to the secondary caps are long and wide and attached to lots of wires. These will take away a lot of heat in a hurry. The 60 watt had no trouble here, though in places I did need to hold the iron in place for several seconds to fully melt the solder.
We are now free to move that pesky mess o' wires away from the secondary side and get a good look at the caps to be replaced. Most of them, anyway. Here we see the three big magnetic amplifier (mag-amp for short) coils and three pi filters that make up the independantly regulated secondary side of the Neo 480W. It is the three pi filters (one each for 3.3v, 5v, and 12v) that do all the noise clean up in here, each has one input cap, one coil (the little black cylinders), and one output cap. The coil has as much import to noise filtering as the caps do, but since we can't really get replacements for them, and they are specially selected according to the rest of the design, we'll just do the caps.
While we're looking at the caps, take note of the height and diameter of the original caps. It is critical to replace these with other caps that will fit, else you'll never get the thing put back together. In this case, we need all 10mm diameters. This was a problem, as replacements in that size for these values can be hard to obtain. I was able to find 10mm replacements for all but the large 4700uF 10v capacitor located on the top left. These 10v caps can be replaced with 6.3v replacements if needed to assist in the diameter problem - they will be located on the 5v and 3.3v rails so they'll still be rated at a high enough voltage to work. Your options on the 12v rail will be limited to 16v caps and over however... if you cannot find anything that will fit, it is acceptable to replace them with lower capacitance parts as long as you don't stray too far from the original value.
For my experiment, I was able to obtain exact replacements save for the 4700uF cap. For this one, I managed to shoehorn in a 12.5mm diameter cap.
Here's the exhaust side of the PCB, where we can see some of the other caps we're going to replace. See the two hiding in the middle just under that big white power resistor? They make up the input and output caps for the 5vsb rail's pi filter. These must be replaced, as they are often the first to go just because they are always filtering power even with the PSU in standby mode. We can see a few other capacitors on the other side of the heatsink we'll be replacing too.
Ok, now comes the fun part - desoldering the caps. Holding the PCB on its side, press gently on one side of the cap you're desoldering while applying the iron to the solder pad beneath. When the first cap lead is free, repeat for the other lead. I've mastered the art of holding up the PCB with one hand while gently prying on the caps with my fingers to desolder them, but if you wish you can use a PCB jig if you have one to hold up the board for you. Or, put it between two heavy objects if you don't mind the risk of static discharge.
Once all caps have been pulled off the board, go back and desolder all the holes using your vacuum bulb, Soldapullt, needle, desoldering station, or what have you. Being a master of the Soldapullt, it took me less than 10 seconds to use it on each through hole.
Nice clean holes, ready for the leads of the new caps. Watch out for surface mount components on the underside when soldering/desoldering - the Antec wasn't so bad, but just wait 'till we get to the FSP, Charlie.
The secondary side with the caps gone. Fortunately, the silkscreening for polarity was correct on this unit. The black shaded area is for the negative lead... this is marked on the capacitors as well, as you can see below.
Here we have 9 pretty Fuhjyyus, all in a row. Not a single one showing outward signs of failure. However, it is possible for these to fail without showing it - it takes an ESR meter to tell. And I don't have one.
The new caps are in place. Most will be held in there tightly enough to stay put while you're soldering them underneath, but some may try to get away. I have a small soldering heatsink I like to clip onto one lead while I solder the other, but I couldn't find it this time. So, I used masking tape to hold the loose caps in place. Try not to bend the leads when soldering... it's not usually harmful to do so, but there is a risk of rupturing the bottom seal in the caps if you do. When that happens, it's dead Jim - replace with another new cap.
New 5vsb caps in place all soldered up. When soldering around the bigger PCB traces and wires, try to hold the iron to the solder pad next to the cap lead alone for several seconds to preheat the solder pad before moving over to the cap lead. This will keep you from cooking the cap with the iron while the trace is still warming up to the correct temperature. When you get every cap soldered in, snip the leads close to the solder joint, and then double check all solder pads to be sure none are bridged with solder where they shouldn't be. Then, resolder the wires to the AC line you pulled off to get the PCB out, and put the bunny, er... PCB back in the box. Darn you and your memorable lines, Mr. Cage.
Some of you may be pointing out that I left one out, labeled C24 on the PCB. To this I respond - there was never a capacitor there. This is because it's for the -5v rail, which is not present on this unit. If you look closely at the heatsink, you'll spot the location of the missing regulator. You can see where the -5v wire gets soldered in next to the missing capacitor.
Now we have one freshly recapped Antec. Get a couple zip ties and clean up the wiring so the fan doesn't eat the pretty multicolored pasghetti, put the cover back on, and get ready to power it up for testing. Make sure you have a DMM or voltmeter handy, you'll need it to check all rails before using this in a computer again.
Power it up like so - a small paper clip or wire between the green wire and any black on the ATX connector. I'm a cheapskate of the worst kind, so I used a segment of cap lead freshly cut from one of the Chemi-Cons used in the recap.
Yay! 5.12 steady volts on the 5 volt rail!
Yay! 12.18 steady volts on the... I hate repetition.
Yes, we're even testing the -12v, -5v (if present), and 5vsb. All rails must be verified. And yes, that fan really is spinning.
Before I show you a bunch of cool scope shots of how the Antec performed before and after the recap, we'll go visit the other patient, the FSP600-80GLC.
Right here on this very site, several versions of the FSP Epsilon design developed an early reputation of having a bit of a ripple and noise issue that saw ripple exceeding ATX specs at full load. I wanted to see just how much capacitor quality had a bearing on this problem. And I use the word "problem" loosely, for it has yet to be proven that up to 200mV of ripple is really that much worse than the usual spec of 120mV. So, I put the unit on the bench and grabbed the same brand and model caps used from the Antec recap, and set to work on one of FSP's latest and greatest designs.
The insides look a bit anemic for a 600 watt power supply, but this isn't your father's 600 watt power supply. It's far more efficient with a higher switching frequency than the designs of yesteryear. You say your father doesn't have any 600 watt power supplies? Hey, my mind's made up, don't confuse me with facts!
Moving on, we need to remove the main PCB just like we did on the Antec. See the brown and blue wires in the below picture? Those are our targets on this unit. Make a cap placement diagram like with the Antec.
Some of the caps we're going to replace. These are made by Capxon, a lesser known but still respected company. You can see one of the screws to remove to get the PCB out - the FSP is a little screwy in how it's mounted in there, bad pun intended.
A close-up of the two wires keeping us from pulling out the PCB. And they're glued in on this one. Gently pry or cut the wires away from the glue so you can desolder them.
The PCB out of the enclosure. Again, we'll ignore the primary side with the single huge cap. The secondary side on the right is where the action is.
That black thing between the big coils and the heatsink is another cap with a sock pulled over it. Don't worry, it's not planning to rob a liquor store, it's just covered up to keep it from heating up too much around those hot little coils and output diode packs. Carefully cut or pry the glue away from all caps.
You may be wondering what's the deal with only two coils in this one. Well, this is a group regulated design, meaning that two rails share a coil. In this case, it's the 5v and 12v rails. While this means loads on one will always affect the other, it is a common practice when space and cost savings are important. Most older designs that didn't cost a mint (and even some that did and still do) use this type of design.
This time, we aren't going to pay too much attention to anything other than the 12v section, for 'tis that area where the ripple is at its worst. But, I ripped out every cap on the secondary and replaced them with Chemi-Con anyway. 12v output is handled by two OST 2200uF caps in parallel, where they are attached to protection circuitry that splits the unit into four 12v rails. Use a good powerful iron again, because this unit has some big traces to solder.
Above we see one of a couple places where soldering the FSP becomes a game of how steady we can keep our irons. One small slip could easily wipe out that little SMD resistor. Soldering on this PSU is not for the newbie I'm afraid.
Old caps gone, never to return. Keep the sock for that one near the heatsink - you'll need it for the new one.
As with the Antec, we fit our new caps in place, holding them in with masking tape if necessary, and then solder them up. Above is a picture of one cap soldered in, ready for lead trimming. Just to the right, you can see the holes of another cap location, and the second place you need to be reeeeeally careful with that iron.
Nice brand new Chemi-Cons all soldered in. KZE and KY used.
Everything back together. Time for testing. Power it on outside a computer as with the Antec. Again, verify all active rails.
3.3v works fine.
12v rail - an oddball measurement like this is typical of a group regulated PSU being run without a load. It's in the ballpark, which is good enough to tell it works.
Here we have the other half of the odd effects of group regulation, a high 5v rail. This is normal. They will even out when the PSU is loaded properly.
Anybody curious to see how they did now? Next page, please. Things are about to get... interesting.
Time for the good stuff, load and ripple testing. The testing was done by jonnyGURU in two phases several weeks apart, using a SunMoon SM8800 ATE and a Stingray USB scope. Loading on both tests was equivalent to 80% of maximum output - this was done to keep the poor Antec from popping in case the caps had already gone to cap heaven.
Here at last are the scope results:
Whaaaaaaa!?!?! Ripple is actually worse now? Wassupwitdat! To be honest, I'm not entirely sure myself. It's likely however that the original Fuhjyyus were chosen specifically for this design by the engineers, and the different characteristics of the United Chemi-Con caps are a contributing factor. One thing I am certain of though is that this PSU now has a longer lifespan thanks to the more heat resistant properties of the Chemi-Con capacitors.
Ok, let's not panic. Maybe the FSP fared better with our little experiment. After all, isn't it the one that needs the most help with ripple?
Oh for the love of... here too the ripple is actually worse after the recap. Why? Good question. It is likely that the new caps are somewhat different performing in circuit than the originals. What can be done about the ripple then? Well, short of a total redesign on the pi filters, nothing. And that redesign must be done at the factory I'm afraid.
My head hurts now. Maybe I should quit pounding it on the desk. Or unplug the soldering iron before I see little dancing bears cavorting across the keyb... uh-oh. Better open a window.
During this little experiment, I hope you gained a measure of how different quality capacitors can affect the operation of a computer power supply. I certainly learned that one cannot always tell how well these will perform simply by what capacitors are used inside them.
If nothing else, I hope you were entertained by my little odyssey into the depths of these two power supplies. Thanks for reading, and do stop by our forums to discuss the article.
Many thanks to the helpful forum members both here and at badcaps.net for the help during the research phase for this article. If you're looking for places to find good capacitors, or just have questions about capacitors in general, badcaps.com is a fantastic resource.