On Sun, 17 Nov 2013, Peter Coghlan wrote:
I have three Alphaserver 1000A machines all of which
have problems.
The bcache failed on all the CPUs - there is a jumper on the CPU card to
disable it and run with reduced performance (one of the SRM memory diagnostics
then fails spectacularly as it divides something by the size of the cache...)
On of the CPUs failed entirely. One of the machines used to randomly halt
and progressed to randomly powering itself off, possibly due to it thinking a
fan is faulty when it is not.
On the rare occasions when I've had one working, I found that the correct
positioning of the CPU in it's slot can be very critical.
I've also had problems with memory failure and found (by doing lots of swapping)
that the SRM can report the wrong memory slot as being the one with the problem.
The power supply sits there getting hot and consuming lots of power when the
front panel power switch is off making it necessary to disconnect it from the
mains when not in use.
I've had so much grief with mine I've lost interest in trying to fix them.
Opinions of the 1000A do not seem to be very high. I think the main
difference between it and the 1000 is that the 1000 has more (E)ISA slots and
the 1000A has more PCI slots. I don't know if they suffer from similar
reliability problems.
Several of my Alphas were developing flaky behavior by the time I
bequeathed them to another collector. I always attributed it to
electromigration issues. That effect was not always modeled well during
the period of time Digital was fabbing semiconductors. My UP2000+ was
progressively losing the ability to drive multiple banks of memory and
wouldn't boot with more than 512M installed at the end.
Interestingly, the low-end units (Multia and PC64) never missed a tick. I
think they were fabbed in a much larger technology.
Steve
--