[GTALUG] War Story: for want of a nail...

[D. Hugh Redelmeier]

(Actually, for want of a nylon post, but that's less poetic.)
Only read this if you are interested in an example of hardware 
debugging.

An older desktop, acting as a MythTV server, started crashing hard.

- would start to reboot, get past POST, but didn't get all the way up.

- didn't seem to be software because the spot that it stopped was not 
  consistent.  It would even crash while I was in grub trying to edit
  the kernel like to show more boot logging.

- Power supplies are often the problem.  My tester said all the voltages 
  were there (but of course that's a pretty coarse test).

- Often contacts stop conducting as well as they used to.  Oxidization?
  Unplugging and plugging in all the RAM, cards, and devices did not fix 
  anything so that probably was not the problem.

- Then I notice the heat sink on the nvidia FeForce 8200 chipset was
  askew.  And something was loose in the bottom of the case.

Here's a picture of the motherboard:
<http://www.ecs.com.tw/ECSWebSite/Product/Product_Detail.aspx?DetailID=873&MenuID=20&LanID=0>
The heatsink in question is orange, near the top right.  There are two
nylon posts that hold the heatsink against the chip (see the white
spots on opposite corners of the heatsink?).

The one post had perished and let go.  Probably slightly explosively
since between the head of the post and the heatsink is a spring (a bit
like the spring in a retractible ballpoint pen).

When I touched the still-functioning post, it broke too.

I figured the computer's behaviour was consistent with the chip overheating.

I didn't know how to source the right nylon replacement posts.
Instead, I used a pair of pairs of cable ties.

I used some 20-year old thermal grease (hey, I seemed to have bought a
lifetime supply back in the day).  I first scraped off the one-time
phase-change thermal pad that had been used.

I was happy that my jury-rigged computer seems to be functioning
correctly now.