Re: BogoMIPS?

Chris Wedgwood (cw@ix.net.nz)
Thu, 5 Feb 1998 18:28:14 +1300


Date: Wed, 4 Feb 1998 16:52:26 -0500 (EST)
From: James Mastros <root@jennifer-unix.dyn.ml.org>
To: Chuck Carson <chuck@digmo.org>
cc: Linux Kernel <linux-kernel@vger.rutgers.edu>, redhat-list@redhat.com
Subject: Re: BogoMIPS?
In-Reply-To: <011501bd3179$761da4e0$4952ce80@chuck.digmo.org>
Message-ID: <Pine.LNX.3.96.980204164550.1205E-100000@jennifer-unix.dyn.ml.org>

[...]

On Wed, 4 Feb 1998, Chuck Carson wrote:
> 1) What is a BogoMIP?

One bogoMIP is one (million?) time(s) around an empty loop in one
second.

> 2) I have noticed that at boot time the system will calculate (?) the
> bogomips for each cpu add their combined total.

The point of bogomips is to be able to run n*bogos to pause for n seconds
exactly. Giving the sum of them is just a feature added to apease those who
use bogomips for bragging rights. (Note that bogomips make very poor
benchmarks: in real use, computers don't spend much time in empty loops).

Some might argue the majority of benchmarks aren't useful for more than just
bragging rights....

I _really_ like lmbench for the simple fact its really useful to actually
seem if some particular optimization or not has made an improvement.

[...]

Pentiums have highly varabile bogomipsage depending on just where the
loop is in memory. When you change the options a kernel is compiled
with, often locations change. It shouldn't matter, so long as the
calculation loop and the work loop have the same cache charactistics.

I think it also depends upon MMX/non-MMX by a _huge_ amount. Presumably the
improved BTB helps here.

-Chris