On 7/10/2016 12:05 AM, Paul wrote:
Bob F wrote:
I played with this a lot a couple years ago. I finally gave up when I
started getting crashes on systems that had worked fine for many
months. I found that the 771 to 775 stickers seemed to fail after long
usage. The system would start crashing, more and more over time. I
finally figured out that if I removed and replaced the sticker, or
removed it and scraped the contacts, then re-installed it, the system
would start work again - for awhile.

I don't know if I had a bad batch of stickers (they were cheap), or if
anyone else has had similar problems. I've got half a dozen of these
processors I could put to use if I came up with a solution.

Has anyone used these stickers over the long term and never had this
problem, or did you have the problem?

I couldn't find anything about this on the "delidded" site that
details this conversion.

http://www.delidded.com/lga-771-to-775-adapter/

The film could be something like this. It has
a relatively high temperature rating, so maybe
the heat of the processor could not be blamed.
Even if the processor body was 100C, this stuff
should hardly even notice.

http://www.dupont.com/content/dam/du...properties.pdf


You have metallurgy and normal force (downward
compression) to consider. The lever on the LGA775
provides some of the compressive force, that makes
the LGA775 contact scheme work.

The LGA775 contacts are springs, with a pointy bit
in the center. The contact is quite sharp. It is
gold plated. So is the flat contact on the processor,
it has a gold plating too.

The spring contact "bites" into the gold on the processor.
That's why the picture on the delidded site, I can tell
that processor has already been through at least one
installation cycle, because the CPU has the bite mark
in the middle of the gold plated contact.

Now, consider what the film is doing. It's a metal to
metal contact, where the film is laid over the processor.
For best results, the "sticker" should be a plated stackup.
Maybe the tracks on the polyimide are copper, and they
should be plated up with nickel, some other metal, then
10u gold. You want a gold on gold contact, because there
is minimal "bite" in this contact. The sticker doesn't have
its own spring with sharp metal point. If the sticker is copper
and is touching gold on the CPU, that's a less ideal solution.
If you're using gold, it should be gold on gold, not
gold on "crap". Gold will not oxidize. 10u gold is so
thin, there is a danger of pinholes.

Where the LGA socket contact approaches the film (and those
two pads), it is going to "bite" into the contact on the film.

An ideal situation would be, if the metallurgy on the film,
used the same stackup as the one on the CPU. I don't know
enough about metallurgy, to guess at the consequences of
getting this wrong. You've probably seen how shiny copper
behaves in air over time, how copper chloride builds up
on exposed copper surfaces and so on. So copper is not
a noble metal, and isn't the best choice for stuff that
is just "rubbing together".

You would have to figure out from your symptoms, whether
the two pads in question are shorted together (LGA775
contact bites all the way through the film, film wasn't
thick enough), or something else is happening, like
an open circuit (where film contacts the CPU).

If those pads are being used for a BSEL mode, then
the processor CPU frequency could be wrong. And maybe
that's a way to detect you're in trouble. Use CPUZ to
check the CPU frequency.

I did a BSEL mod on my Asrock board, placing an insulator
on one of the BSEL pins, then soldering a wire to the back
of the PCB on the LGA775 socket. Using the wire, I could
set the FSB to run at FSB800 or FSB1066. I didn't overclock
for long, because it wasn't stable (my particular CPU),
but the insulation selected and the wire, have not malfunctioned
with time. I avoid taking the CPU out of that socket, as
it's a bit tricky to get the insulation in just the right
position. Unlike your Xeon, no really messy mod is needed,
just insulate a pin in the socket, then drive the floating
signal on the socket that results, with a logic level.
All I need to do is ground the wire, to force FSB1066
speed. Because the motherboard had no VCore adjustment,
I also have a booster mod on VCore, to boost the voltage
by 0.1V above normal. The VCore regulator chip was an
excellent choice by Asrock, and even though they were
cheap *******s and didn't give a BIOS setting, you
can still get out your soldering iron and "fix it" :-)

Good luck on your engineering failure mode analysis.

Paul

Thanks for the analysis. This mod for 771 to 775 conversion seems to
involve swapping the connections at 2 pins.

I guess what I really would like to know is, are there better stickers
that don't have a problem with failures, and if so, how do I find those
ones? Or, maybe I found the only bad supplier and I could try anyone else?