[geeks] weird power supply behaviour

[Mouse]

>>> The PS is willnig to power the MB and both disks, but not MB and
>>> disk B alone.  [...]
>> [...]
> [...] when I next have the leisure to experiment with this and write
> back with my results.

Okay, I've now played with it a bit.

I've carefully restricted myself to two of the disk power connectors on
the power supply, calling them C and D (A and B I'm still using for the
disks).  A/C, A/D, B/C, and B/D mean "disk [A or B] on PS connector [C
or D].  With a * appended, means the data cable is connected to the
disk (I haven't tried data cables to both disks).  A/- and B/- mean
power is not connected to that disk.  C and D are not connected outside
the PS case - that is, they use separate wires on the bundle of wires
emerging from the PS housing.

Disk A is a Seagate ST34311A; disk B is a Western Digital 64AA.

My first set of tests, in which I tested the 15 cases not involving the
data cable connected to an unpowered disk, showed a simple and perfect
correlation: if disk B was powered, it failed; if not, it worked.

This made me start to doubt my own report, so I started just cycling
power with both disks connected, to no avail.

Then I pulled the MB power connector, manually grounding power-on, and
found that all combinations worked.  Plugging the MB back in, I found
myself back in the weird state: MB A/- B/C and MB A/- B/D both fail,
but MB A/C B/D and MB B/C A/D both work.  I'm now unable to make it
fail with both disks connected.

Switching power supply connectors (that is, switching C and D) has made
no difference in any of my tests, regardless of what they are plugged
into or not.

I don't think the "too little load" theory is tenable, since it works
fine with the disks and no MB.

I'm not sure why it would be desriable to leave the MB running
memtest86 or some such; all of these tests powered down before the BIOS
came up far enough to do much of anything with the RAM.  Still, on the
(highly likely) theory that you people know more than I do about x86, I
hooked up a CD drive (PS connector C, not that I expect that to matter)
and started booting CDs until I found one with memtest86 on it.  It's
labeled "Finnix 101 Build 2473" in my handwriting, so presumably that's
what it is; when running, memtest86 displays "Memtest86+ v4.10" in
black-on-green except that the + is red on green blinking, either
blinking + to space, or blinking red-on-green to green-on-green.

In case it matters, here's how memtest86 describes the machine:

Celeron 398.7MHz
L1 Cache:   16K   3726 MB/s
L2 Cache:  128K   1011 MB/s
L3 Cache:       None
Memory  :  511M    177 MB/s
Chipset : Intel i810E

The chipset detection appears to be correct; there is a chip visible
marked with the Intel logo and

	Intel. 810
	FW82810E
	L005IG01
	SL3MD
	INTELM) '98     <-- the M on this line is actually circled,
			    rather like the ) symbol but with the C
			    swapped out for an M.
	E9491811IB
	   005 KOREA

It's no speed demon; it's been going for 16 minutes but is only 24% of
the way through the first pass (both self-reported).  I'll leave it
overnight or until it reports trouble - or until you people come up
with a better suggestion. :)

/~\ The ASCII				  Mouse
\ / Ribbon Campaign
 X  Against HTML		mouse at rodents-montreal.org
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B