Paul Heinlein <[email protected]> writes:
> On Tue, 2 Nov 2010, Rich Shepard wrote:
>> On Sun, 31 Oct 2010, Bill Barry wrote:
>>
>>> I should add that this seems to say that memtest86 crashed not that it
>>> reported you have bad memory.
>>
>> Certainly looks that way.
>>
>> I replaced memtest86 with memtest86+. It ran all night (was in the 13th
>> pass when I stopped it). In the first pass 8 errors were detected; no
>> additional ones found.
>>
>> What is considered an acceptable error number? The 8 found here seems high
>> to me, but I know the areas are very tightly packed on the silicon.
>>
>> Should I swap these for a new pair?
>
> IMO, even a single memory error means the underlying DIMM (or whatever)
> should be swapped out. memtest86+ should run completely clean.
*nod* This is not like a hard disk, where they automatically swap out those
bad bits on write or anything.
It is also notable that memtest86+ is good, but usually not as hard as running
real software on the machine is; we have previously had occasional faulty
sticks of memory that would run clean for weeks under memtest86+, but would
oops within ten minutes on corruption in the Linux MM lists.[1]
Daniel
Footnotes:
[1] At repeatable locations, even, which surprised me. Presumably something
in the access patterns was the trigger.
--
✣ Daniel Pittman ✉ [email protected] ☎ +61 401 155 707
♽ made with 100 percent post-consumer electrons
_______________________________________________
PLUG mailing list
[email protected]
http://lists.pdxlinux.org/mailman/listinfo/plug
