P4 overheating very likely due to misaligned duct

[R Clint Whaley <rwhaley@cs.utk.edu>]

Guys,

I have still not proven anything, in that I have not taken the machine
into a controlled temp. environment.  However, I have not been able to
make the overheat occur since I opened the case and fixed the ductwork.
Jack asked me to try another experiment: see if purposely messing with
the duct will cause an overheat.

So, first I took the case apart, and misaligned the duct so it looked 
roughly like I remember it being when I opened the case when we first
had overheating problems.  With this configuration, I could still not
cause an overheat.  So, I removed the ductwork completely, and the
overheating happened just as we had experienced before, where you
can cause it effectively at will.  Notice it happened as before:
when you begin running the computationally intensive code you get
good performance (in this case, 4Gflop), but in 5-10 min it has
dropped significantly (say to around 2.5Gflop).

So, we haven't proven anything, but given that no amount of trying with
a correctly installed duct has achieved an overheat (and I let the program
that was overheating in 7Min run for a 8 hours trying to cause it), and
that we can reproduce the behavior by yanking the duct, it does seem
likely that that was the problem.

So, if you see overheating on a P4, my advice is to crack the case and make
sure everything is in order . . .

Cheers,
Clint