Intel Guy wrote:
I have several idential Soyo i845 motherboards that date from 2003 -
2005 time-frame, and several 2.53 ghz Pentium-4 CPU's (.13 micron single
core) with 8k/512k cache.

All boards have had various electrolytic capacitors replaced within the
past year.

One board in particular has been problematic over the past few months
despite replacing most of it's capacitors.

The system (running Windoze) performs spontaneous reboots and even
though a mem-test might take over 50 passes after 6 hours of testing, it
will show memory errors at some point.

But the memtest shows something interesting.

When run on two similar board-CPU systems, memtest says this:

L1 cache - 8kb - 20763 mb/sec
L2 cache - 512kb - 17714 mb/sec
Memory - 512mb - 1052 mb/sec


But on the problem system, I get this:

L1 cache - 8kb - 87xx mb/sec
L2 cache - 512kb - 65xx mb/sec
Memory - 512mb - 559 mb/sec

I used "xx" because I didn't write it down.

But clearly there is something strange about those numbers. Bios
settings were the same for all boards, ram setting set to SPD.

I pulled what now could be a problem CPU and tried it in one of the
other motherboards (a board that has a 2.6 ghz celeron) and got the same
low score for cache/ram speed.

I replaced the now likely questionable CPU on what I thought was a
problem board with the next best available one that I had on-hand (3.2
ghz Celeron-D) and got these scores:

L1 cache - 16kb - 22376 mb/sec
L2 cache - 256kb - 19630 mb/sec
Memory - 512mb - 1033 mb/sec

So the board and memory can indeed perform similarly to the other
problem-free boards, but I now seem to have a P4 CPU that has internal
issues with regard to cache and memory speed (the only parameters I'm
easily able to test).

This CPU is SL6EG.

Anyone know how a CPU could partially fail along these lines?

You haven't told us if these CPUs have hyperthreading, and if so if it's
enabled. I have never tested telling a BIOS to enable HT if the CPU didn't
support it, but it comes to mind as one of the things which used to
differentiate between systems I used when the P4 was current tech.

I suspect the CPU is in some way defective (might even be one of the ones with
the F00F bug Intel replaced for free), but I do remember HT being an issue, and
at least with Linux a feature which was good for about 30% better performance
for some real work.