Post by Mayayana
| Well guess what?
| I swapped out the two 1GB Ram cards for two spares I had. Said PC has
| been on for some 72 hours continuously even with Avast installed. Has
| yet to shut down. Problem solved.
You might also try putting them back. I've run into
problems with RAM before, taken them out one at
a time to find the culprit, then eventually ended up
with all of them going back, concluding that at least
one of them was a tiny bit loose. A good, tight
connection is *so* important with the parts inside
the case. And moving a machine in transport can
loosen those connections.
The thickness of multi-layer PCBs, isn't controlled that tightly.
When the press operator inspects their laminate collection, they
have to pick and choose laminates, to get the correct overall
dimension. On controlled impedance products, there isn't a lot
of maneuvering room when it comes to thickness (each layer
thickness also has to be controlled, to meet the measured
electrical impedance target). A "test coupon" on the edge
of a larger PCB assembly, is where a TDR is connected to
measure the impedance. The test coupon is a means to accept
or reject a PCB lot, for electrical performance.
We had to reject one shipment of finished blanks at work,
because they'd managed to go more than 10% over the
specified thickness (too fat!). They can also err in the other
direction, and make things too thin. Usually, a too-thin
item, it's a "purposeful" error. The person making the
product (blank), knows what they're doing, and they've done
this when compromising on some other requirement.
Fabrication requires the operator of the press, to
trade all the requirements, and stay within the tolerance
envelope, so the customer won't be pissed off with them.
I've seen DIMMs like that, and the surface finish and
reflectivity tells me non-standard (unusual) material
choices were made. Maybe the dielectric constant was
quite different or something. And this is still a fiberglass
and resin based product, but something is different about it.
We had a local representative, who would come in, and in
a two hour lecture, attempt to explain all this stuff.
But it's pretty hard to cover everything the guy knows, in
two hours. So he's only able to give a rough feel for
the job he does. Basically, when you send him a design,
that dude is adjusting *everything* :-( No aspect of
the job you send him, is "untouched". Everything is
translated, and fed through a $5000 piece of software,
to meet the spec, rather than the actual data in the
design file shipped. Very unsettling. If the plot file
said to make a copper track 5 mils wide, he'd adjust it and
make it 5.2 mils wide. You'd need a microscope to
see these corrections. He does an ERC check to make
sure no modification he's made, caused a short circuit.
I had one of these twits, *erase* something they noticed
in a design. Now, normally I would reward such a person
with a "good catch" verbal award. Except, he didn't phone
me. He didn't warn me. That there was a short circuit in the
color keys... It looked like a photo-plotter error. A second
company, upon receiving the design, phoned me immediately
and warned me. And sure enough, I could see the plotter error.
(The plotter uses a light, an aperture wheel, and "flashes"
light patterns on to a photo-resist. The wrong aperture
had been selected by the software.)
So when you get a DIMM or a motherboard, that something
is out of the ordinary, "wee willy adjuster" could've done it.
I don't consider this a good engineering workflow, but
what are you going to do ? They're all doing this,
so you cannot punish all of them for messing around.
We did actually put some of these shops on the banned list.
It all depended on what sort of grief they caused.
The edge of the DIMM, the finish has varied from
one generation to another.
Ones shaped like this (seen on DDR2 and DDR3 here)
those are likely to be making poor contact because they're
not actually fully seated. It's difficult to get a satisfying
"click" from the latches with those. Those hurt your
fingers to install, especially when the heat spreader
doesn't have an area suitable for fingers to push on.
So maybe one end is riding high.
The older ones, the edge-card had a finish like this.
This is considered "optional" in terms of a JEDEC spec.
They don't have to do this. This involves more process
steps, although with NC machines, it's just time on the
machine. They have to mill a profile, put the keying
slot on the thing, so some amount of NC is still required.
The square edge in the above one, can be covered by milling
the entire edge.
The contacts in the socket, don't have any spring action
to them. (Not like the contacts in an RJ-45.) There's no
compliance to them that I can see. And the finish
on the top of the contact, when mixed with the squared
DIMM profile (the top one), had better deflect the DIMM
so it doesn't catch on a contact. That's called
"connector capture" - the design has to encourage the
items to mate, so they don't bind and stuff gets
broken. Mechanical engineers do overall tolerance analysis,
to make sure stuff like that works.
It's been a long time, since I saw an actual "bend" to
a contact in a memory slot. Nothing like that has happened
to me recently. The bent pin (a slight bend, not a 90
degree failure like in a male IDE connector), was long enough
ago, I can't even tell you what generation of RAM that was.
Could have been SDRAM, or FPM/EDO or earlier, it was that long ago.
I can't see how you could bend something like that, back
into place. At least, if the contacts are as stiff as
the modern ones are. It's bound to continue to
stick out and get in the way.