Subject: Re: Port performance comparison?
To: David Maxwell <>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
List: current-users
Date: 08/26/1997 21:29:25
On Wed, 27 Aug 1997 00:58:10 -0300 (ADT), (David Maxwell) writes:

>Has anyone ever compiled a chart comparing NetBSD's performance on the
>different ports?

>(Please keep in mind I've never used some of this hardware, and some only
>with native OS)

>                Alpha
>                PowerPC
>                Sparc
>                arm32
>                i386
>                pmax
>                hp300
>mac68k - mvme68k - x68k - amiga - atari - sun3
>                pc532
>                pica
>                vax

>Anyone have corrections to my guestimate?

The basic problem is that the range within each CPU family is too wide
for this to be meaningful.

Consider that a single CPU implementation will have a 2x-3x
performance range through improved process technology and/or shrinks,
without changing the basic implementation. Then consider 4 or 5
implementations per CPu family.

the total performance range is wide enough that a simple linear
ranking just isn't meaningful, unless you state clearly how you are
computing rankings, and why.  And many rankings are possible.  See
below for details.

* Do you rank a CPU family by where its current top-end implementation
  falls, compared to the top-end of other CPU families?
  Or by where where it falls against contemporary implementations?

  For example, m68ks were faster than contemorary i386es, at least until
  the 040 and then the 060 was late,  and 486es overtook m68ks.
  Yet current Pentiums and PPro cores are clearly faster than any m68k.
  Which is the correct ranking, esp. for comparing late-80s boxes
  against other late-80s boxes?

* if memory serves, mips-based boxes were faster than comparable
  Sparc-based  boxes from the late 80s/early 90s.

* Current top-end mips systems are comparable to sparcs. NetBSD
  doesn't support either mips-IV or sun4u.  Yet you list sparc above i386.
  Which is the faster system, a supersparc, or a 300MHz Pentium-II?

* To a first approximation, an mc680x0 is an  m680x0.  Memory system
  timing, cache,  and I/O subsystems would dominate.   sun3s
  did better on those metrics until the 50Mhz 030 HPollo boxes (hp400) 
  came out. And  HPollo stopped development of 040 systems long before
  040 CPU development stopped;  Mac 040 systems (quadras) are
  faster than HP300s , even if the average  030 mac isn't.
  And HP300s with HP-IB disks were significantly slower than sun3s 
  from the same period with similar CPUs. 

  etc., etc.,

* A Pica with a mips3 CPU should be faster than r3000 based pmaxes :).