17 Nov
2013
17 Nov
'13
6:58 p.m.
(I think I asked this once before, but I can't remember the result and I can't
find the thread in the archive)
I have an Alphaserver 1000 that I would like to resurrect. It periodically
hangs or reboots. When it reboots the error is usually "Machine check while
in palcode". When it hangs, the hang is so severe the halt button does not
work. Either way, it always happens within 20 minutes of booting. I ran all
of the diagnostics you get by moving the CPU card jumper (cache memory tests,
RAM tests) and they run for an hour or so with no error. (I am assuming the
diagnostic halts on error instead of just looping again, is that correct?)
During the course of these I moved the jumper to the wrong place and wiped
out SRM, so I had to reload SRM from floppy using the failsafe loader. That
worked. The SRM diagnostics will also run for as long as I want without
error. Does anyone have any idea what's wrong here or is there some kind of
XXDP-like thing I can run that will better test the machine?
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.
Regards,
Peter Coghlan.
18 Nov
18 Nov
8:24 a.m.
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
--
10:01 a.m.
On Nov 18, 2013, at 8:24 AM, Steven Hirsch <snhirsch at gmail.com> wrote:
On Sun, 17 Nov 2013, Peter Coghlan wrote:
> I have three Alphaserver 1000A machines all of which have problems.
...
> 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.
I have a 1000 here and it works fine. My problem has been one of finding
disks that don't conk out quickly.
On the 1000A, the additional PCI slots (after the 3 primary ones) are
behind a bridge, so you lose some performance (mostly due to latency) on
them and SRM doesn't always know how to probe boot devices on the
secondary bus. All the EISA slots on both models are on the same bus
(after a PCI-EISA bridge, of course).
My 2100A has an interesting issue, though. It wouldn't boot with the
512MB board that I received it with (earlier this year), but when I
bought 3 more 512MB boards and stuck the original one at the end, it
boots happily with all 4. Go figure.
If anyone has a 5/266 board for the 2100/2100A that they're willing to
part with for less than an arm and a leg, though, I'd like to make this
an SMP machine.
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.
Yeah, electromigration is a known problem, especially with earlier
Alphas (can't remember if it plagued 21064 or 21164 more). I hope mine
last long enough for me to have fun with them.
- Dave
3848
days inactive
3849
days old
2 comments
3 participants
participants (3)
-
cctech＠beyondthepale.ie -
fraveydank＠gmail.com -
snhirsch＠gmail.com