Here at CMU, departments routinely dispose of obsolete or damaged hardware. They drop it in hallways, and it disappears. The unwritten rule is: If it is in the hallway it is free, unless marked "NOT TRASH" or "DO NOT TAKE".
Occasionally, my friends and I go on long scavenging trips where we walk every hall in the main Computer Science building and sometimes others. However, we found a dedicated disposal area, which we now frequent daily. Too much free time I guess! Finds include MANY 386, 486, and socket 5 / 7 based systems, network cards, video cards, monitors, fans, cases, etc. The older hardware is usually in fine condition, but newer stuff tends to have issues, or be marked bad. A couple weeks ago, there was a stack of 12 power supplies marked bad -- 8 of which work fine. Two currently power my computers. There was also an 8gig hard drive marked "cannot read". It had a linux partition on it. It works fine. Go figure. The most recent (and coolest) find was two Slot-A motherboards and a Slot 1 motherboard. The Slot-A's weren't marked, the Slot 1 had a physically-damaged dimm slot. Warhorse took them all, as I wasn't especially interested (stupidly assuming they were useless). We have not yet looked at the Slot 1 motherboard.
How I Got the Board
Warhorse tested the two Slot A motherboards and found that one locked up during POST, the other during boot. Neither were anywhere near useless. Being the kind soul that I am, I traded him a fully functional Socket 8 (PPro) motherboard and a dead pentium I had opened. (Both were part of the collection of stuff scavenged in the past).
Why the Board was Trash
After looking at the board for a few seconds, I realized what the problem was -- 16 capacitors, presumably part of the CPU voltage regulator were damaged (on both boards). If you take a look at the pics, you'll see some brown stuff on some of the caps. That's not supposed to be there. Also, though it is difficult to see (if possible at all), the tops of the 16 caps are all convex (dented outward). A normal cap is level.
A Stupid Idea
Fortunately, earlier in the year, I had built two arrays of capacitors for the sole purpoes of making a big spark (I'm easily amused). The original 16 caps were all 2200uF, rated for 10V.
After looking at the traces, I noticed that 10 of the caps were definitely wired in parallel, and the other 6 were paired. Because of silkscreens,it was too hard to tell if all 16 were in parallel. To play it safe, I stuck with the original 4-bank setup (1x10, 3x2). I desoldered all 16 capacitors -- and ran into a problem.
A Problem & Solution
The new caps wouldn't fit through the existing holes, and the existing solder was difficult to melt, so I whipped out -- the dremel and the smallest drill bit I could find. It made quick work of the holes, although it was as wide as the trace on one side, meaning that the trace was completely cut. I also had to take care not to hit nearby traces.
The next thing to do was figure out exactly which caps to use. I decided on a 10mF 16V cap, a 4.7mF 35V, a 3.3mF 25V, and a 10V 3.9mF for the bank of 10, a 3900uF (10V) cap for one of the banks of 2, and a 4700uF (6.3V) cap for each of the other two banks. I wasn't sure how well the 6.3V caps would work, so I tested them with the 12V rail of my power supply (in series with a resistor of course) and they didn't explode. Of course, that doesn't mean they won't blow up given some time. I'm fairly sure they are safe to use though, since I believe CPUs are powered from the 5 volt rail, so I would not expect to encounter voltages much higher than 5 volts, if that. I have not measured the actual voltage to see if it is 5, 3.3, vcore or something else (I think I should do that).
Soldering the capacitors in was somewhat difficult, since with almost no exposed metal (as a result of my drilling), the solder has less to stick to. After jury-rigging the 10mF and 4.7mF caps and getting everything together, I was ready to test.
I chose to use my slot tbird 700 over my athlon 550 (650 core, stable at 750 or more) since the tbird is a poor overclocker and I expeceted the athlon to do 800+ if I moved to water cooling. The computer did not post using a known-bad stick of ram and an unknown video card, so I switched to a known-good PCI video card. Still no post, so I hooked up a speaker. The beep code turned out to be bad memory (the ram is KNOWN BAD/unreliable) -- so I swapped in one of my 128MB sticks. It posted, but my cpu cooling was not working well (you try holding a p90 socket-7 heatsink to the core of a ~35watt heat source for any period of time ;) ) and I had to shut down. I switched to holding a slot A heatsink against the core, and booted successfully into linux, from which I cleanly shut down. Note that before my "modification" the computer locked up during post. Success? Couldn't tell yet. Warhorse lent me some ram and a CPU with heatsink for testing (since I need my ram in my real computer) and I couldn't get it to post. Evidently, I'm retarded. Since neither of my athlons have their plastic cases, I didn't remember how much force an athlon normally takes to insert. After properly seating it, the system booted. I ran genome@home and disconnected the monitor and keyboard. Unfortunately, at some point it crashed. I probably knocked the video card loose though, since when I restarted it the computer continued to run for a few hours under full load.
Results & Conclusion
I am now certain that the repais were successful. However, the solder really stuck poorly to the metal there was, and broke prettty quickly. To compensate, I exposed more metal by sanding the green insulating layer of the mobo on the traces of interest. The solder sticks very well to the exposed areas. A glue gun added the finishing touches. The caps are firmly held in place and the solders are all good. The motherboard can handle 80-watt CPUs (overclocked athlon 550) so the voltage regulator looks like it is doing its job just fine.
Unfortunately, it doesn't look like the AGP slot works. However, all PCI slots work fine, and I had experienced problems with two of them before resoldering the caps with the exposed traces. I guess they're somehow related.
Why did I name the box jesus.res? It's back from the dead, and lazarus.res was already taken. Is this offensive? Don't know, don't care ;). If you've got a better idea for a name, let me know.
Pics of the process
note: this was written about a year and a half ago. I'm no longer using any Slot A Athlons.