FreeBSD Crashes with AMD

[Brian F. Feldman]

Bakul Shah <bakul at BitBlocks.com> wrote:
> > I just thought my idea was pretty cute, it would also be nice to say to people
> > with mystery SIGSEGV's that the break into the loader type 'memtest' and see 
> > if they get errors :)
> 
> It was creative alright!
> 
> My experience has been that memtest like tests do not help
> with nasty, marginal power/timing related errors that only up
> on a heavily loaded multiuser os.  Start a few compiles,
> finds, pure number crunching programs, throw in a few crashme
> kind of tests and see how well things stand up.  Then run the
> same load at highest/lowest rated temperatures and for 24
> hours or more.

Commonly referred to as "Prime95" in the Windows/OCers world, the Great 
Internet Mersennes Prime Search client is a very good indicator of system 
stability.  It has a generic "burn in" mode you can run and a mode that 
actually searches for primes; it can be found in ports/math/mprime.  I run 
an instance for each CPU on my A7M266-D (w/ECC), and it's pretty damn stable:

  PID USERNAME PRI NICE   SIZE    RES STATE  C   TIME   WCPU    CPU COMMAND
55725 green    155   36 17372K 15104K RUN    0 606.1H 78.91% 78.91% mprime-real
27850 green    155   36 14532K 12228K CPU1   0 662.7H 76.81% 76.81% mprime-real

-- 
Brian Fundakowski Feldman                           \'[ FreeBSD ]''''''''''\
  <> green at FreeBSD.org                               \  The Power to Serve! \
 Opinions expressed are my own.                       \,,,,,,,,,,,,,,,,,,,,,,\