Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 25 Apr 1995 14:54:57 +1000
From:      Bruce Evans <bde@zeta.org.au>
To:        agl@mac.glas.apc.org, jgreco@brasil.moneng.mei.com
Cc:        freebsd-hackers@FreeBSD.org
Subject:   Re: Is P5 is faster than DX4 100 under FreeBSD?
Message-ID:  <199504250454.OAA15990@godzilla.zeta.org.au>

next in thread | raw e-mail | index | archive | help
>> Linux FAQ (or was it HOWTO?) states that bogomips are measured as
>> DX2/DX4 Mhz * 0.5
>> P5      Mhz * 0.39
>...
>Now, the 386/40 with Linux would be about 7.9 bogomips.  The 386/40 with
>FreeBSD would be about 8.6 bogomips (a little over the predicted 8?)...  the
>486dx2/80 would be 35.1 (a little under the predicted 40?), and the Pentium
>would be 117.0, which does not seem reasonable, although the P90 does appear
>to be at least 2x on most of my tests over a 486dx2/80...  so perhaps the
>P90 performs very well (and/or optimizes much better) on tight loops.
>Interestingly enough, the Linux formula you gave would suggest that the P90
>would be rated at 35.1 bogomips.

The bogimips test happened to have instructions that pipelined very
poorly for Pentiums.  This can't be fixed because then the test would be
useless for historical comparisons.  This doesn't matter because the
test is bogus.  On all x86's with an on-chip cache (i.e., for everything
except i386's), the loop in the test takes a certain number of cycles
that depends only on the cpu, so its results are given very accurately
by the formula

	bogomips = speed_in_MHz / cycles_per_test / fudge_factor(= 2?)

cycles_per_test is easy to calculate from the best-case instruction
timings.

>I then ftp'd a Linux boot disk and did some measurements on the machines I
>had easy access to:

>machine		bogo
>386dx/40	 8.0
>486dx2/80	41.1
>pentium 90	36.2

For i386's the speed depends on the external cache and is difficult to
predict.

Bruce



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199504250454.OAA15990>