Capacitors in Computers

[davmax]

The Role of Capacitors in Computer Power Supplies



Introduction



This small note is for those that like to modify their motherboards to improve performance or reliability. Hopefully it will improve understanding.



Definitions

• VRM = Voltage Regulator Module
• ESR = Equivalent Series Resistance
• ESL = Equivalent Series Inductance
• CPU = Central Processor Unit
• Mhz = millions of cycles per second

Computer power supplies including the VRM (for CPU) are all closed loop switching voltage regulators. At the output of each an electrolytic capacitor plays some important roles in conjunction with an output filter inductance.



The output inductance carries out the role of current smoothing and this assists with ripple voltage reduction. The output capacitor in the case of a VRM is usually a number of capacitors connected in parallel.



Electrolytic output capacitor roles.

• The capacitor/s function as part of the regulation of the DC output in that they affect the regulator response time. These regulators respond relatively slowly to change (in milliseconds) this is largely controlled by the value of output capacitance. With no output capacitance there would be many problems (even with no load) including a fast response that would create instability. Increasing the capacitance will lengthen the regulator response time. In this role the capacitor helps maintain stability of regulation.
• The capacitance is an energy store (bulk storage) that delivers transient energy to the load where the speed required to respond is outside the capability of the regulator. Later the regulator tops up the capacitor. ESR is important in that a high ESR will cause an excessive voltage drop during a transient load.
• The regulator does not respond to the switching speeds of the power MOSFETs. The output capacitance together with the series inductance form a filter to remove ripple frequencies. ESR is important in bypassing the ripple current so that ripple at the load is minimal.

It is clear that capacitor ESR plays an important role dealing with the high frequencies outside of the regulator response time. There are harmonics of the switching frequency that are higher than an electrolytic can handle with ESR. Fortunately very little, if any, gets past the output inductance.



It is unlikely that ceramic capacitors will be found in a typical computer power supply or VRM because the power supply filtering method is sufficient. Ceramic capacitors are found on the supply lines at the CPU; they are applied to supply the very fast supply transient needs of the CPU and must be placed very close to the CPU socket to avoid ESL problems.



What the impacts of replacing capacitors?

• Increasing the value of output capacitance will slightly slow the regulator response and will improve the energy storage.
• Using capacitors with lower ESR. This will not normally affect the regulator response or stability. Here it should be understood that ESR plays important roles.
  • It must be lower than the load resistance and to draw the bulk of ripple current.
  • Being a resistance it acts as a means of removing the ripple energy in terms of heat. If the ESR was reduced to zero there would be no way of removing the energy, if this were possible a near lossless output filter would have what is termed a high Q factor and damped oscillation is likely to occur (the ripple energy needs to be absorbed), this would be at a frequency outside of the regulator control. It is therefore possible that if ESR is reduced to a very low value voltage spikes will develop on the regulator output. This would be critical in a VRM. Output capacitance total ESR should remain at an effective ratio to the load resistance. See the following Ratio of Load to ESR example.

Ratio of Load to ESR example.



CPU power is 60 watts at 1.6 volts (V core). Calculated load resistance E2/W = 42.6 milliohm

Six Capacitors in parallel @ 15 milliohm each = 2.5 milliohm

A ratio of 17: 1

This means the load will receive 5.88% of the available supply ripple.



Now change the capacitors to 9 milliohm each. Result is 1.5 milliohm

The ratio is now 28:1

Load now receives 3.57% ripple.

The absolute value of the ripple voltage available depends on the power supply components and switch frequency.



What about adding ceramic capacitors?



Great care must be taken in that ceramic capacitors have a very low ESR and if any ESL is present in the motherboard traces a high Q circuit can be established eg the power line from the VRM to the CPU can create a very small ESL. With a ceramic at the CPU there is no problem, however if a ceramic is connected at the output of the VRM (in parallel with an electrolytic capacitor) a high Q circuit can be formed by the ESL between the two ceramics, this high Q will magnify a small very high frequency signal of the same frequency as the resonant frequency formed by these components. This does not happen with electrolytic capacitors at the output because at these very high frequencies they are lossy (a low Q factor).



Motherboard manufacturers can control the value of stray capacitance and ESL to suit a particular need. Most often they try to minimize both. Sometimes ESL may be introduced to improve current sharing in parallel connected devices.



Adding ceramics to the output of the main computer power supply has been tried and found to produce spikes. Here a high Q capacitor is reacting with inductance and there is plenty in the power supply leads.



If improvement is to be tried the best place is at the CPU. An increase in ceramic capacitance may help.



Closing comment



All the above relates to the current and common types of switching power regulators in computers. New technology may emerge that operates at very high switching speeds (Mhz range). The designers claim that response is so fast that large electrolytic bulk storage capacitors are not needed, ceramics will be used.

[Topcat]

Nice read Dave, thanks.

Stickied.

[Big Pope]

Great, interest. Thanks, dav.

[trodas]

Okay, let's try another example. A Barton at 200x13 - 2600Mhz will have exactly 100W maximum. 100 / 1.8 = 55.5A going to the poor CPU, witch means it's load resistance is 0.032 ohms now.

The DFI LP B got 5 caps. Originaly there was a 5x 3300uF KZG ones. Their ESL is rated as 0.012 - but what is their ESR? Why you used the ESL values there, when the R = U/I could serve there giving us (6.3/2.8A ripple 100kHz) 2.25 ohms?

If we use the ESL (not R), then we ending up with 0.0024 miliohms, and that is ratio 1:13.3x. World overclocking records for AXP clock are made on THIS exactly.
(or we go with the calculated ESR to reach 0.45 ohms, giving the ratio reversed as 14:1)


I replaced these KZG with Panny FM 4700uF 6.3V, ending up at ESL 0.013 per piece, 0.0026 whole and therefore the ratio is 1:12.3. Quite similar, eh? And the calculated ESR ratio is there 1.735 ohms as they can give 3.63 Ampers in 6.3V, ending up at 0.34 ohms whole, witch is reversed 10,6:1 ratio.


With excessive coolin I reached almost 3Ghz clock (before it topped at around 2700) and 2869Mhz is even recorded and CPU-Z verified there:
http://valid.x86-secret.com/show_oc?id=84643

YET under normal cooling (just one CPU fan) I having troubles running at 2300Mhz stable, while before I was running watercooled and fanless at 2600Mhz all the day and stable.

I starting to feel that increasing the imput Vcore regulator caps aren't good idea at all, yet the rule you give us is the lower the ESR, the better - and just don't use the ceramics. I learned that too, yet the record was made with ceramics all over the place. Some places seems to be even more sensitive for ceramics that others.
For example I had no trouble of 2869 (I was at 2900 as well) speeds with 100nF ceramics behind every of the 5 end Vcore caps, but as soon as I add ceramics to the bottom of every Vcore imput filter caps, the machine stop posting untill I removed them...

Now I learn that ceramics are bad for VRM, so I removed it all, yet still 2300Mhz is not stable anymore and I need just 1.575Vcore for that and honestly I never run my AXP-M 2600+ THAT low in THIS board. I always stated with 2500Mhz as start of overclocking, as the absolute bottom... And like I say, 2600Mhz worked beautifully...

I think the high influence have these input filter caps as well. Increasing their capacity and specs seems VERY dangerous for stability. Dunno why.

Input
1500uF KZG 16V - 2500mA ripple/0.013 ESL - very good, 2600Mhz stable
2200uF FM 16V - 3190mA ripple/0.015 ESL - 2300 Mhz semistable
1500uF GC 16V - 2900mA ripple/0.010 ESL - to be determined

Output
3300uF KZG 6.3V - 2800mA ripple/0.012 ESL - very good, 2600Mhz stable
4700uF FM 6.3V - 3630mA ripple/0.013 ESL - 2300 Mhz semistable
3300uF GA 6.3V - 4690mA ripple/0.0065 ESL - to be determined

So, according to you, these GA should be optimal, right? And what do you suggest in the pre-Vcore VRM filter caps? There is clearly options - I could use also:
2200uF GC 16V - 3370mA ripple/0.008 ESL
or
1500uF GD 16V - 2550mA ripple/0.013 ESL

...and these GD are VERY similar in characteristic to the used KZG ones! Same ESL as well, as almost exactly same ripple. Now witch one could yield best results any why? That is the puzzling question to me... ...and maybe I would rather hear some theoretic thoughts before I start soldering and testing. The aim is, of course, bring 2600Mhz back to life and stable.

edit:
PS. the overclocking problems and stability are NOT HEAT RELATED, as I use good aircooling and even the moving air can now cool down the components nearby a little, instead of no cooling them before...

[davmax]

Trodas,

I am not sure why you are quoting ESL for electro caps. The reason I do NOT is that ESL would only become a significant in the caps if the power supply frequency was increased. Increasing frequency would increase ESL this would make ripple removal harder because ESL would increase the impedance Z within the capacitor reducing the ripple through the capacitor and more in the load (CPU).

You quote an ESL for KZG 3300u as 0.012. This is incorrect. UCC quote 0.012 impedance Z which is a composite of ESR+ESL+Xc. At the frequencies applied in current power supplies the ESR is dominant ie ESL and Xc can be ignored for practical purposes. So take the impedance Z value as ESR.

The your R=2.25ohms is not correct. There is not a 6.3 volt ripple. The correct way to look at this is E(ripple) = 0.012 x 2.8 =33.6 millivolts. The ripple voltage is quite low to create a 2.8 amp current flow.

Concerning ceramic caps. It entirely depends on the motherboard characteristics(particularly ESL) and the frequencies involved as to whether success will be had by adding ceramics at the VRM output. Yes it is likely you can be successful. This emphasis is on BE CAREFUL because where high Q components exist there is always a risk of a resonant frequency causing an unwanted effect. Certainly the safest place for these caps is at the CPU socket.

Increasing frequency of the CPU will improve the Xc performance of the ceramic but the ESR will remain the same. Xc is the capacitive impedance that decreases with frequency increase.

Looking at your tables of input and output caps I cannot tell you why you get the results you have seen. There does not seem to be any logical relationship. I can say as I have before that the ESR (not ESL) of the input caps does not need to be as low as that of the output caps. The input caps see a higher load value therefore the ESR to load ratio is changed.

A simple calculation for 100W where the CPU load for output caps is 0.032ohms. The load seen by the input caps for a 5 volt supply = (25/100)x0.9 = 0.23ohms for a 12 volt input = (144/100)x0.9 =1.3 ohms

Hope this helps

[davmax]

Trodas. Re Samxon GA. In theory they would certainly bypass more ripple. The best way is to try. If ripple is already within spec for the CPU you may gain nothing. Questions are: Will this be reducing ESR too low? Will this improve tranient response or will ringing (spikes) appear?

[davmax]

I just realised I had fallen into a small trap. We have been so busy talking about the ESL effect that it was forgottem the ESL is a value of inductance, not impedance that is correctly termed reactance Xl.

So the impedance Z of a capacitor is a composite of R (ESR) + Xl + Xc

Xl = 2Pi x f x L where f =frequency and L =inductance

ESL is an inductance.

It should also be realised that the values of Xc and Xl do not add directly to R because each is effectively 90 degres out of phase with the value represented by R (ESR). Xc lags and Xl leads so they are in fact opposing values. One will subtract from the other. At a resonant frequency Xc = Xl so they cancel out leaving only ESR and the lower the value of ESR the higher the Q value and the more ringing or damped oscillation.

[Tom41]

Just a small footnote here. You may have noticed that large capacitors, especially in PSUs, have a plastic cover on the top. At first I thought this was foolish, as it was covering the vents and could lead to violent failure.

In fact, the plastic is there to provide electrical insulation. Big caps like these are frequently fed with mains voltage or higher, and the top of the cap will be at this high voltage. If you touch the top, or if something conductive comes into contact with a high-voltage cap like that, you'll get a severe electric shock!

[zaniac1998]

Trodus is right

[TELVM]

Originally Posted by Tom41

Just a small footnote here. You may have noticed that large capacitors, especially in PSUs, have a plastic cover on the top. At first I thought this was foolish, as it was covering the vents and could lead to violent failure.

In fact, the plastic is there to provide electrical insulation. Big caps like these are frequently fed with mains voltage or higher, and the top of the cap will be at this high voltage. If you touch the top, or if something conductive comes into contact with a high-voltage cap like that, you'll get a severe electric shock!

Ouch, these Panasonic EE-TS came bald:

[endooo]

iam new in electronics and i have a few question about capacitors, can u test capacitor ESR value in circuit (computer motherboard or PSU)? because i tested some "bulge" capacitor on my motherboard using ESR70 and they showing low ESR but leaky capacitance, when i pull them all, it shows high ESR, is there any components in motherboard that affects the way ESR reads?, do bad caps generates so high capacitance than there normal/ original capacitance value? because i tested one 1000uf 16v (not in circuit) using my ESR70 it give me 0.06 ESR but 2,400 capacitance

[brethin]

You Can Not test capacitor ESR in circuit and get a accurate reading most of the time.

[endooo]

thanks brethin i just thought peak atlast ESR70 can test caps in-circuit, is it possible that bad caps can generate times two than its original value? because ive tested 1000uf caps off the board, and it shows 2400 uf, how it is possible?

[momaka]

Yes, it's possible to get 2x capacitance (or sometimes even more) when the capacitor develops high internal leakage. Last week I pulled a bunch of GSC caps from an old socket A motherboard. They all looked fine and ESR measured okay. However, the capacitance on most the 1000 uF ones read 1500-2500 uF. Obviously that's bad

Defective Nichicon HM and HN capacitors from year 2001 to 2004 fail in this way too most of the time. At least they bulge, though.

[mariushm]

Motherboards often have several capacitors in parallel, to reduce ESR and capacitance.

When you're reading 2400uF you may actually measure the capacitance of 2 or 3 capacitors, not just one.

Atlas ESR70 and other meters "can" measure in circuit, but that just means the voltage used to test is low enough not to activate stuff around capacitors like transistors, mosfets etc.

"Can measure" does not equal "correctly measures", you have to determine if it's possible or not to measure the capacitor in circuit.

This being said, like momaka says, it's possible for a capacitor out of circuit to show higher capacitance.

[ftechno]

hi i want replace ost rlx 470uf 16v with 470uf 16v SANYO SEPC Aluminum solid Capacitor but the difference is in Rated ripple current (mArms) is it possible

[momaka]

@ ftechno:
If it's on a motherboard, it should be OK.
In a PSU, it's a different story.

[danyboy]

Good Read!! Thanks.

[flaworlds]

Very helpful. Thanks

[t333]

good information, thanks