How is bogomips computed?

Chet Nichols III chet.nichols at gmail.com
Tue Jul 22 00:18:34 UTC 2008


hey yong-
i'd suggest disabling hyperthreading. it might "look" good, but in reality,
you probably wont see any performance benefits at all- and this would be one
reason. i'm not sure if the process that computes bogomips runs across all
the processors in parallel, but if it does, it would explain why your AMD
architecture looks funky (since there's not going to be a full 100%
available cpu available for it). try disabling hyperthreading and checking
out the result for bogomips, i'm interested to see what it says :D

chet

On Fri, Jul 11, 2008 at 3:00 PM, Yong Huang <yong321 at yahoo.com> wrote:

> Hi, Chet,
>
> I think the ht flag in /proc/cpuinfo indicates whether I have
> hyperthreading.
> So yes, I have it:
>
> $ grep -w ht /proc/cpuinfo
> flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca
> cmov
> pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext lm 3dnowext 3dnow
> pni
> cx16
> ...
>
> On a different server with exactly same OS release and kernel version and
> also
> ht flag in cpuinfo, all CPUs' bogomips are the same:
>
> $ grep bogomips /proc/cpuinfo
> bogomips        : 6675.48
> bogomips        : 6666.32
> bogomips        : 6666.20
> bogomips        : 6666.40
>
> However, I think this is very relevant. The server that shows different
> bogomips uses AuthenticAMD CPUs while the second server I mentioned uses
> GenuineIntel. I checked two other boxes, one AMD one Intel, and found the
> "problem" on AMD only.
>
> Katsu, I read the full article you suggested. It doesnt' seem to answer my
> question.
>
> I think I can assume AMD CPUs generally run slower with the same CPU
> rating, if
> we test many types of applications and compute an average.
>
> Yong Huang
>
>
> *********** Chet Nichols III wrote ***********
>
> Do you happen to have hyperthreading enabled?
>
> On Wed, Jul 9, 2008 at 1:23 PM, Yong Huang <yong321 at yahoo.com> wrote:
>
> > Even though bogomips as shown in /proc/cpuinfo is a rough estimation of
> CPU
> > capability, there must be some truth to it. Is it computed by doing a
> > simple
> > busy loop to see how many iterations it can go through within some set
> > time? I
> > can't imagine half of the CPUs on one of our boxes have only half the
> > bogomips
> > of the others:
> >
> > $ grep bogomips /proc/cpuinfo
> > bogomips        : 4018.62
> > bogomips        : 2009.31
> > bogomips        : 4018.62
> > bogomips        : 2009.31
> > bogomips        : 4018.62
> > bogomips        : 2009.31
> > bogomips        : 4018.62
> > bogomips        : 2009.31
> >
> > Yong Huang
>
>
>
>
> --
> redhat-list mailing list
> unsubscribe mailto:redhat-list-request at redhat.com?subject=unsubscribe
> https://www.redhat.com/mailman/listinfo/redhat-list
>



-- 
----------------------------------------
chet nichols III
chet.nichols at gmail.com
aim: chet / twitter: chet
http://chetnichols.org
----------------------------------------



More information about the redhat-list mailing list