[ltp] Stress testing for undervolting

TNKS linux-thinkpad@linux-thinkpad.org
Sat, 15 Jul 2006 10:27:19 -0500


I installed the Linux-PHC patch to undervolt my Thinkpad T42.  I also
downloaded the MPrime binary to test for prime numbers and see how low of a
voltage my machine could tolerate.  

The wiki's on-line have some really good content about how to perform such
stress testing, but I was maybe hoping for a bit more "live" of a

I know that there are two camps of people, those that will tolerate a bad
calculation, and those that won't.  I'm definitely the later.  So
unfortunately, that obligates me to some pretty long-duration stress
testing.  It's very easy to find a voltage cliff, where the CPU coughs
almost immediately (within a few seconds).  If I bump up the voltage by
16mV (the smallest interval the patch allows me to step) from that point,
the problem seems to dramatically improve.  The longest test I've run to
date has only been about a half an hour, but it's been long enough to prove
that 16mV from the cliff isn't enough of a margin away from the cliff.  

Documentation on-line recommends 24 hours of testing for a proposed
undervoltage setting.  Does anyone have an opinion on this?  Has anyone
seen a first failure at, say, the 22nd hour of stress testing?  I just
don't have a feel for statistics of this stress test.  I know there's a
risk in undervolting, and I'll never have the same confidence as with not
undervolting.  But how much confidence do you guys feel 24 hours of testing
is really worth?

Also, for Linux (I'm not really in the mood to fuzt around with somehow
getting Windows on this machine), are there any other good tests to try
instead of mprime?  

Thanks for your input,