Recapping an Epox 8KRA2+

Recapping a dead motherboard – Super Nade

Flailboy, one our members, approached me with a request to revive his dead motherboard. Having performed surgery on several motherboards before, I was running through the possibilities on what might have gone wrong. The usual suspect is the infamous “bad caps” phenomenon, where the electrolytic capacitors on a motherboard fail catastrophically. Analyzing modes of failure would entail an article in itself and so we will limit ourselves to a visual after the fact.

The board arrived and this is what I saw:

The board was filled with capacitors leaking their electrolytes (henceforth referred to as bad caps). Note that the caps that decouple VDIMM, near the chipset, near the AGP connector and the ones near the ATX connector were completely decimated. The bad caps are circled in red.

Now that we have seen how bad the situation is, I’ll list out the capacitors on this board with reference to the above picture:

1. CPU VRM Side (Blue): Teapo 1500uF , 6.3V ; Sanyo 3300uF, 6.3V ; GSC 1000uF, 6.3V
2. VDimm Side (Red): GSC 1500uF, 6.3V ; GSC 1000uF 6.3V
3. Chipset and AGP side: GSC 1000uF, 6.3V ; GSC 1500uF 6.3V

Of the lot, only Sanyo has a good reputation in the tech community. There have been a few reports of dodgy Teapos before, but I have not heard any recent reports of failure. I’d mark Teapo as an average brand. GSCs, on the other hand, are clearly crap. After seeing many boards with dead GSC’s, I’m not too surprised.

So the task is clear cut it seems – replace the dead capacitors and the job is done! Is it that simple? The answer is a Yes and a No. Replacing capacitors can be done with a bit of practice, but the more important issue is to understand what the replacements need to be.

There are two ways to go about this process and I’ll try to explain both techniques. Note that everything in this article is based on my experience and ideas gleaned from reading technical documentation. I am not a power electronics engineer, so there may be inaccuracies in this write-up. If you spot anything amiss, please bring them to my attention!

1. Replace the bad caps with ones having the exact same capacitance and voltage rating:
   This is the easiest way to do things if you need to get the job done without worrying about the deeper technical details involved. Just pick equivalents from a reliable manufacturer (see the appendix for a list of good brands and series).

2. Use a replacement that is not of the same kind (i.e. aqueous electrolyte) or value, but is better suited for the application.

In the spirit of OCForums, this is going to be my preferred approach. To perform this operation, we must understand the function of the bulk capacitance on the motherboard. On the CPU VRM side, bulk capacitance is necessary to filter out transients in the power delivery mechanism. This usually manifests itself as fluctuations or “ripple” in the inductor current.

On a motherboard, clean power to the CPU is ensured by filtering out AC components that arise due to the switching nature of the delivery scheme. There are mainly two kinds of filtering. The first kind uses bulk electrolytic capacitors as a shunt (or low impedance path) to get rid of low frequency fluctuations in inductor current (about 300 kHz). The second set of filters is located in the CPU socket or on the CPU itself. These consist of Multi-Layer Ceramic Capacitors (MLCCs) which have low impedance over a wide band of frequencies.

We are working with the transient suppressing capacitors here. The prime considerations when finding a replacement are (a) It should be able to handle large amplitude of ripple current (b) It should have low ESR.

To make a choice I decided to understand the nuts and bolts of CPU power delivery and the best place to find such information is with the CPU Manufacturer. My reference guide for this article was the AMD Tech Doc as given HERE.

Let us assume the worst case scenario of a whopping 10A transient at a CPU Voltage of 3.2V (PER PHASE!!!) as given by example 2, page 23. In this context, let us look at the existing layout of the VRM output.

We have 5 x 1500 uF Teapo, 1 x 1000uF GSC and 3 x 3300 uF Sanyo making it a grand total of 18400 uF. These caps are in parallel. Assuming an ESR of 40 mOhm per cap, we have the net ESR as 4.45 mOhm. Following example two, we arrive at a figure of 1800 uF per phase for a total of 5400 uF overall.

Remember that if capacitors are in parallel, the net ESR is lower than the least value of ESR in the parallel network. If this is the rationale, we can understand why Epox used 3 good low ESR SANYO, 5 mediocre Teapo and one horrendous GSC. The answer is to keep costs down. Sadly, this has come back to haunt them with the GSC exploding prematurely.

Now let us look at the replacement candidates. All Teapos and GSCs were replaced by Sanyo OSCON SEPC 560uF, 4V radial caps. These caps have an ultra-low ESR of 7m Ohm! Now we have a capacitive network of 3 x 3300 uF Sanyo Electrolytic and 6 x 560 uF of Sanyo OS-CON caps. The grand total comes to 13260 uF.

Assuming the 3300 uF cans have an ESR of 40 mOhm, the equivalent ESR for the network is 1.072 mOhm. Therefore the required bulk capacitance is approximately 1120 uF per phase or 3360 uF. So, we are well over the minimum limit by an order of magnitude.

So, what is the big deal? After all, both before and after scenarios are above the capacitance budget. Are we missing something here? The answer is YES. The point of this exercise so far, was to support my conjecture that bulk capacitance values per se are not critical.

The real check that needs to be used is to calculate the ripple current per phase. Assuming that the original network has a ripple current of 1660 mA @ 25 C per cap and that there are 3 caps per phase, each must support 1667 mA (note that we are dealing with 3 caps in parallel).

Each OSCON has a ripple current rating of 6100 mA and this is well above the maximum permitted ripple current. Wrapping up the analysis, the conclusions can be summarized as follows:

There are other aspects like ESL and its influence on VDroop that make OS-CON like capacitors very attractive but that is beyond the scope of this article.
{mospagebreak}

The Original Setup:

Note that this is a 3 phase design. This can also serve as a classic example of why ripple currents in SMPS power delivery can be rather harmful. Look at this naïve example: Say your SMPS puts out an AC ripple of 250mV @ 100 kHz. The ripple current without taking ESL into account would be 6250 mA (ESL will reduce this number)! Note that although ESL seems useful here, it plays a very detrimental role in other important scenarios.

Removed the crapacitors. Replacements are United Chemicon LXZ, KY and Sanyo OS-CON SEPC.

I’m using a copper BNC cable to clean the holes up. It is very easy to do if you have a good iron. The important thing to remember is to use more than a dab of solder to remove the bad caps. Solder to solder heat transfer is rapid because of the 2D spread of molten solder. Try to dab the back side of the dead cap hole with solder and it will come off easily. Using additional solder makes cleaning the hole easy because if the hole is completely filled, melting it when you insert the needle is painless and the result is shown in the next picture.

Hole nice and clean!

Replacements near the AGP port.

Memtest stable for 22 hours after replacing only the dead caps.

VRM Caps have been replaced. The Sanyo Electrolytics have been left untouched.

The finale – Memtest stable for 37 hours. I hope you have enjoyed reading this article as much as I enjoyed fixing dead motherboards!

Leave a message on My Blog to check for updates in analysis of this board or to get in touch with me.

The author is not responsible for anything (incidental or consequential) that happens to you, your dog, neighbor, mom, girlfriend… etc if you choose to tinker with electronics. The author strongly discourages amateurs, idiots and other sundry noobs from operating a soldering iron or handling sensitive electronics. If you do blow something up, do send me a postcard!

Super Nade @ OCForums