Part 2: The Repair

To get warmed up, I just replaced the caps. This was pretty straight forward. My solder sucker was utterly inadequate for the task, but heating the leads and wiggling the old caps out was another method I had seen. Followed by trimming and tinning the leads of the new caps and wiggling them back in as you heat the leads one and then the other. I found trimming the leads to different lengths helped get one lead in solidly before moving onto the other. I had issues with the first popping out or not being as seated as I thought when I began trying to wiggle in the second.

Here are the offending leaky caps.

I replaced every 6.3V 3300uF cap I saw, since I think all but one were leaking. The board had about 20 6.3V 1000uF caps though, and only two looked bad. And they were two closest to obvious heat sources, the AGP slot and the chipset. So I only replaced those.

Moving onto the traces, this part had the steapest learning curve. I studied this video as well as I could. But it just wasn't sufficient. His wire stayed put far better than mine ever did. I think he was using 34 gauge wire, and mine was closer to 40 gauge. The labels were all in Chinese, and all I could make out was 0.1mm. So my wire was squirrely as fuck. Even the heat from my soldering iron caused it to arc away. The first night absolutely nothing worked and I gave up frustrated. The precision tweezers had too small of a gripping surface and the wire constantly squirmed out of them or went off in all kinds of directions. And the wire never, ever, bonded to anything. Eventually it would get tangled or stuck to the iron. I grabbed a failure beer from the outside fridge and went back into the house in defeat.

The second night, I tried again. I did some more reading, and the wire I'm using is likely enameled. I say "likely" because as I said before, the label is in Chinese. So I scraped all around the wire with an exacto knife to get it off and/or rough up the wire so it solders better. Then I tinned it. Then I tinned the traces. I used larger tweezers with a wider gripping surface. I used a far more precise soldering tip, despite the guy in the video using a fairly fat one he appears to practically paint solder onto the wire with. I had some success. I got 2 out of the 7 traces done. I grabbed a victory beer from the outside fridge and strutted back into the house.

The third night I felt good. I decided to try using the broad tip I saw used in the video. The first thing it did was cause the two traces I did the night before to stick to it and come off the board. Fuck. I cooled the iron down with some water, and immediately switched tips back to the one I was using before. I was getting frustrated. I sipped some whisky to keep my hands from shaking. I'm not sure it helped.

After this setback, my approach this night was to work from the inside out, using small pieces of masking tape to get the wire in approximately the correct place. Then gently nudge it with tweezers, or sometimes my finger, into the correct position. Then I would press it down with the precision tip onto the trace. Lots of flux, no additional solder. The tinning on the leads and the wire was sufficient. The iron was tinned mostly to help conduct heat. After a bit of a rough start, I made my way through all 7 leads. The ones closest to the chipset came last when I'd had the most practice. By the end I found the precision tweezers came in handy after all. I'd use them to kind of bracket the wire while the iron came down on top of it, all at the same point so it had nowhere to go. I'd then work my way along the trace like that, but having to move quickly lest the wire heat up too much and release itself from the solder. Copper does that apparently.

I cleaned everything up with some isopropyl alcohol. I finished off my whisky, and nervously came back to the house. I shakily put the computer back together, and had my wife come hold my hand as goofy as that sounds. I was incredibly nervous. I took a deep breath, hit the power button and...

I went on to get Windows 2000 installed, DirectX 9c, nforce3 drivers, Geforce drivers and audio drivers. Curiously enough, when the nforce package installed it's GART driver, the display corrupted. Uninstalling it appeared to have no negative effects. Either the trace repair wasn't as good as I thought it was, the drivers are fucky, or the chipset cooked a little bit when the previous owner had the fan die.

It ran Return to Castle Wolfenstein great, and got a pretty good score in 3DMark2001 in line with what would be expected.

All that's left is to put some solder mask on the repair and this board is finished.

Part 3: How well does it work really

So things seemed OK with a Geforce 4800 without any GART driver installed. Wolfenstein ran OK, 3DMark2001 got a score in the ballpark of what I'd expect. But this is an Athlon 64 system. I don't want to run it with a Geforce 4!

I actually have another Athlon 64 system with a Geforce 6800 GT. I've been meaning to find a backup for the GF 6800 case it dies it's final death on me. Once upon a time, it looked like it was having some memory issues, and lacking any better way to fix it, I heat gunned the memory. It actually worked, brought the card back to life, and it's been in use intermittently for about a year since. I know, I know, but my choices were immediately shell out for a replacement, or make a ghetto repair. Sourcing memory chips and replacing them was beyond my means and skill. Still is. My options were throw it away, or give the heat gun a shot. But I know it's days are numbered all the same.

Anyways, I finally won an auction for a Geforce 7800 GS, one of the appropriate back ups I had in mind. The market is so fucked. You see resellers often putting them up for $100-150, and yet I won an auction with the only bid from a normal dude at $50. Never buy from resellers.

So I pop 7800 GS onto the test bench, run 3DMark05 and... 2000 3D Marks. That doesn't sound right. I'm expecting something in the 5000-6000 range. I run GPU-Z and it turns out I'm running at PCI speeds over the AGP socket on account of not having the GART driver installed.

Yeah, that makes sense.

On a lark I try the GART driver again, and the display corrupts again. I remove the GART driver. I actually swap GPUs from my existing A64 system. Putting the Geforce 6800 GT on the test bench still yields about 2000 3D Marks, and installing the GART driver still corrupts the display. Meanwhile, in my other, undamaged A64 system the 7800 GS is scoring about 6000 3D Marks. So definitely not a graphics card issue.

Eventually I had the idea to drop the AGP speed from 8x to 4x in the bios. This actually fixed all my issues when the damaged board is running the GF 6800. Unfortunately, the GF 7800 still has issues. Presumably the trace repair introduced an unacceptable amount of signal noise/delay for the AGP slot to run at full speed. And maybe the AGP-PCIe bridge on the GF 7800 compounds that somehow versus the GF 6800. Who knows! Not me!

But for now, I have a mostly viable Athlon 64 System running the Geforce 6800 GT, scoring about 5000 in 3DMark05, which is about where it should be. I'm sure I'm losing some bleeding edge performance in AGP 4x mode. But I always recall the purported bandwidth of contemporary AGP or PCIe specs being overkill for contemporary expansion cards. Mostly I'm just happy my first trace repair went as well as it did, even if it's not perfect.