Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!usc!apple!tahoe!olympus!steve
From: steve@olympus.wrcr.unr.edu (Stephen Wheatcraft)
Newsgroups: comp.graphics
Subject: Re: Personal IRIS
Keywords: price personal IRIS
Message-ID: <4299@tahoe.unr.edu>
Date: 1 Aug 90 20:56:24 GMT
References: <850@earth.cs.utexas.edu> <7009@helios.TAMU.EDU>
Sender: news@tahoe.unr.edu
Reply-To: steve@olympus.UUCP (Stephen Wheatcraft)
Organization: DRI-WRC Reno
Lines: 68

In article <7009@helios.TAMU.EDU> gordon@cs.tamu.edu (Dan Gordon) writes:
>The IBM RISC-based 6000 workstations apparently have the same
>graphics engines as the SGI. However, their processors seem to
>be much faster (according to figures provided by IBM). Does 
>anybody have any data that compares these machines with the
>various Silicon Graphics workstations? Please post or email to
>
>gordon@cs.tamu.edu
>
>Thanks - Dan Gordon.

I have been doing some benchmark comparisons of IBM
R-6000, Sun 3/80, Sun 3/280 and Sillicon Graphics
Personal Iris 4-D25TG. I don't have the figures for the
SGI machines, but here are the numbers for the others:

Sun 3/80 w/ 68881 - non-optimized code
      100.8 real        97.4 user         0.6 sys  
      339.0 real       332.4 user         0.9 sys  
       52.4 real        49.7 user         0.7 sys  
479.8u 3.5s 8:15 97% 0+416k 8+75io 33pf+0w

Sun 3/280 w/fpa - non-optimized code
       81.6 real        47.4 user         2.6 sys  
      209.7 real       195.0 user         1.6 sys  
       38.5 real        35.6 user         0.5 sys  
278.3u 5.4s 5:32 85% 0+416k 18+114io 6pf+0w

IBM Risc 6000 320 - optimized code
	6.0 real         5.9 user         0.0 sys
       10.0 real         9.7 user         0.2 sys
        8.5 real         8.2 user         0.1 sys
Total:
       25.85s real      24.07s user       0.63s sys

Ratios:

Sun 3-80/IBM = 479.8/25.85 = 18.6
Sun 3-280/IBM = 278.3/25.85 = 10.8

The reason that the Sun benchmarks were run using
non-optimized code was that the Sun compiler will
not optimize my benchmark program.
The IBM compiler was able to optimize it.

The benchmark is a set of three programs:
1. 2-d fast fourier transform (128x128 points)
2. multi-grid fluid flow (in porous media) solver
3. particle tracking solute transport in porous media

The SGI people are currently running the same programs on
a SG PI 4D-25TG. I will post the final results here in a
few weeks (after I get back from vacation).

As you can see, the speed up of the IBM relative to the
Sun machines is quite impressive, better than their
published mips ratio (27.5/3). Part of the reason is that
the Sun compiler would not optimize the second program
(for reasons I won't get into!). Nevertheless, these are
real numbers, real applications, not pac benchmarks.


Steve Wheatcraft                Internet: steve@olympus.wrc.unr.edu
Desert Research Institute                 (134.197.1.70)
University of Nevada System     Bitnet  : steve@unsvax.bitnet
P.O. Box 60220                  AT&T    : (702) 673-7393
Reno, NV 89506                  FAX     : (702) 673-7397