Re: CPU load testing
> On Sat, May 11, 2002 at 09:08:55PM -0500, Kirk Strauser wrote:
> | At 2002-05-11T23:43:31Z, Scott Henson <email@example.com> writes:
> | > On Sat, 2002-05-11 at 16:37, Sven Hoexter wrote:
> | >> I prefer cat /dev/urandom > /dev/null
> | > Dont hose your system entropy. Try
> | >
> | > cat /dev/zero > /dev/null
> | >
> | > same result but you still have your entropy.
> | Both of those would get the CPU nice and hot, but if I'm not mistaken,
> | neither one would really exercise the FPUs or memory busses.
> | Wouldn't a better test be, say, a looped Quake timedemo in
> | software-3D mode?
> This will likely test your video card, if you have a decent one
> (unlike mine).
> | Or perhaps Seti-at-home (doesn't it use FP vs. integer math?) run
> | multiple times?
> That sounds like a good idea.
> If you want to test the memory bus, use memtest86. It's designed for
> that :-). (or compile a kernel, that tends to identify bad memory
> when you see gcc segfault)
Thanks to all who wrote back - lots of good ideas!
I've just discovered something new which points the finger at the net card /
network even more - when doing "ifconfig" during a copying of lots of files
over the network, I noticed the "collisions:" reading for eth0 was in the
thousands and going up constantly! A look at my hub confirmed this: the
"collision" light was flashing constantly, which assumedly shouldn't be
happening. (The other two machines can talk for ages over the network and you
never get collisions.)
To UNSUBSCRIBE, email to firstname.lastname@example.org
with a subject of "unsubscribe". Trouble? Contact email@example.com