I've tried something that seems to have an immediate effect. Turning down
the overclocking. I had been running the 2500+ (Athlon mobile, 266Mhz
Barton) at close to 3200+ speeds
(195x12). It had gotten bad to the point that even when I simply turned on
the
computer, the bios screen was corrupted. I turned it down to 200x11 and
that stopped. However, waking from standby is still an issue. I'm hoping
that it's just been the
effect of overclocking it for nearly a year has begun to take its toll on
the cpu.
I'm wondering what has the most damaging effect on the cpu, frequency or
multiplier.
Neither. The CPU is quite likely to overclock to the same
speed it could all along, rather there might be other
variables such as contacts degrading (socket or memory),
motherboard power (VRM subcircuit) degraded or power supply
itself degrading.
However, AFAIK you haven't menitoned the particulars about
the overclocking, like what voltage you were running. If
you were putting 2.1V through it then needing to use
peltiers to keep it from popping, that may have a worse
long-term effect than more modest o'c methods. Having a few
mobile Bartons here, at least two are XP2400-2500 and
overclocked past 2.3GHz but if I recall corrrectly, only
needed modest < 1.75V vcore to get there.
Waking from standby is even more likely a power issue, if
the CPU were effected at all as that is a sudden surge,
large change of state.
Would it be less damaging to run at 183x13 as opposed to
200x11? (both give about 2.2Ghz).
I think your math is a bit off, 183 x 13 is not "about
2.2GHz", it's 2.38GHz. Some Mobile Bartons will do 2.38GHz
fine, others need very high vcore to do it, some need very
high-end cooling to do it, but some just won't do it at all,
stably. By stable i mean that you can fire up Prime95
Torture Test and run the Large In-Place FFTs test for
several hours with 0 errors. Not only will this check the
calculations but put the thermal stress on it.
Along this line of thinking, you should remove the heatsink,
clean off old thermal compound and reapply it. If you had
been using silicone based compound (assume any unknown
compound that doesn't specify it's composition, is silicone
based) then degraded thermal compound could be one problem,
though probably not the only problem as it wouldn't effect
turning on or coming out of a sleep state where the CPU was
cool at the time.
Depends on what is wrong. Generarlly speaking if your
memory contacts are good/clean, if your northbridge is well
cooled, you should retain stability at the higher FSB speed
and be able to use it. However, all else being equal,
adding these non-CPU variables in means that inevitably if
"X" # of MHz is stable, it'd be ever-moreso at the lower FSB
and higher multiplier.
Or is it strictly the speed that's the
most taxing? I guess I'll just keep tuning it down and see if there's a
point where the issue stops.
Depends on what the specific problem is. If it's power, the
total MHz and the vcore needed to get there are most
sigificant. However, power IS regulated down to supply the
northbridge and memory too, so if those subcircircuits are
effected, you also have need to lower their speeds (lower
FSB and memory bus) to regain their stability.
In other words, if you had a new motherboard and/or new PSU
of the same quality (since different motherboards will vary
+- a few % on how far they can o'c a given part) you would
most likely be able to run that CPU at same speed you had
all along.
All of this is rather beside the point though, that because
you have the ability to adjust these settings you should do
so and chech which has an effect.
Set the FSB and memory bus to 200 again (Or whatever it was
at previously as o'c) and test it with memtest86 (Or
memtest86+) while the multiplier is adjusted lower, like
around 8X such that it's so low you effectively isolate the
FSB... goal NOT being to see if CPU is stable but rather to
remove it from the equation altogether and only have the
stress on the FSB and memory for the memtest testing.
If memtest runs ok for several hours, lower the FSB and
memory bus to (always keep memory bus set to synchronous,
100% of FSB), then run memtest86 again. The reason for
this is that when you lower the memory bus the bios will
automagically set tighter memory timings. The alternative
to the 2nd run of memtest would be to observe the specific
timings which passed testing at 200MHz memory bus and
manually setting same timings at the lowered memory bus
speed.
Does your board even allow multiplier settings over 12.5X?
I mean, does it toggle the 5th multipler bit or will setting
13X just result in 5X?
You might also examine your board for failing capacitors,
and touch-test them while running the Prime95 torture test-
if any around the CPU feel more than moderately warm, you
should consider lowering the vcore a little then whatever
CPU speed reduction is necessary to retain stability at the
lower vcore. That is, assuming you want to prolong the life
of the motherboard. If the capacitors are already bulging
though, it's a bit too late for that.
You have made multiple changes to your system, you might
consider reverting back to the prior configuration then
retesting after making each change in turn (referring back
to the opening power about power, memory and video changes).
Too many variables at this point really, as mentioned
throughout my post you need to isolate these systematically
to find the fault. I'm sure I've overlooked a few things
but this post is already too long so I leave it as-is till
you try a few things.
I'm not so sure your need is to just reduce CPU speed to "X"
# of Mhz, except maybe within context of a degrading
motherboard or if the power supply is struggling, BUT if I
were forced to choose, I'd keep the FSB higher if it'll stay
stable.
