SUN vs SGI

Rob Harper Rob.Harper at csc.fi
Tue May 11 06:46:31 EST 1993


The discussion on SUN vs SGI was debated from a very personal
viewpoint. Here are some hard facts gleaned from another newsgroup that
discuss things like price/performance. It is worth a read. Standard
disclaimers apply.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% CLIP %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

I'd like to thank the over ten respondents who provided information related
to my recent posting looking for relative benchmarks and other information
comparing SGI Indigo machines to Sun SPARCstations. I will summarize the
responses briefly below.
 
Most useful (from a statistics point of view) is a posting I read in the
comp.sys.sun.hardware newsgroup (included next).  Can anyone augment this
table with (credible) data for the (R3000- and R4000-based) Indigo machines?
 
The way it looks here (and in Michael Schulman's note below it), the difference
between the mid-range SPARCstations and Indigos really isn't all that big.
(I guess I'll upgrade my SPARCstations-IPX to a SparcStation 10/30 then.)
 
============================================================================
        From: jacobsd at jasper.CS.ORST.EDU (Dana Jacobsen)
        Newsgroups: comp.arch,comp.sys.sun.hardware
        Subject: Re: SPECrate for various vendors
        Summary: SPECrate table
        Date: 8 May 93 03:30:32 GMT
        Organization: Computer Science Department, Oregon State University
 
Here is a table of all the SPECrate numbers I've seen so far.  There's a
little summary at the end.  If anyone has any other numbers or corrections,
please send them to me.  It is a little hard to keep track of these
if one doesn't get the SPEC newsletter (it's way too expensive for me to
get, and the people at work seem to think we do OK by grabbing numbers from
publications, so they won't pay for it either).  The numbers tend to
change through time (as new OS's and compilers get used), as well as being
subject to mistake (often the only way to get a number is from a competitors
brief, and they tend to use older numbers or misplace digits (the DEC briefs
say that the IBM RS/6000-580 has a SPECint92 of 61.7, even though IBM
repeatedly states in it's documents that it has a 67.1 rating).  Another
problem is when a vendor shifts compilers and one number goes up and the
other goes down.  Which should I use?  <sigh So far the SPECrate numbers
have been subject to far less change than the SPEC92 numbers.
 
*****************************************************************************
  SPEC92rate benchmark results
*****************************************************************************
 
Man    Model                    Procs   Sr92int   Sr92fp     Approx list price
------------------------------------------------------------------------------
Sun    SparcStation 2                       517      541        15,000
Sun    SparcServer 600-140          4     1,847    1,930
 
Sun    SparcStation Classic                 626      498.1       6,000
Sun    SparcStation LX                      626      498.1       9,000
 
Sun    SparcStation 10/30LC               1,072    1,282        16,000
Sun    SparcStation 10/30                 1,072    1,282        19,000
Sun    SparcStation 10/40                 1,191    1,427        21,000
Sun    SparcStation 10/41                 1,264    1,607        23,000
Sun    SparcStation 10/51                 1,847    1,930        28,000
Sun    SparcStation 10/402MP        2     2,112    2,378        25,000
Sun    SparcStation 10/412MP        2     2,411    2,854
Sun    SparcStation 10/512MP        2     2,950    3,744        34,000
Sun    SparcStation 10/54           4     5,063    6,142        46,000
 
Sun    SparcCenter 2000             2     2,326    2,953        95,000
Sun    SparcCenter 2000             4     4,539    5,627       130,000
Sun    SparcCenter 2000             8     8,047   10,600       197,000
 
 
DEC    3000/300L AXP   (OSF/1)            1,081    1,480         7,300
DEC    3000/300 AXP    (OSF/1)            1,535    2,137        10,700
DEC    3000/400 AXP                       1,549    2,631        16,000
DEC    3000/400 AXP    (OSF/1)            1,763    2,662        16,000
DEC    3000/500 AXP                       1,762    2,967        39,000
DEC    3000/500 AXP    (OSF/1)            1,997    3,023        39,000
 
DEC    4000/610 AXP                       1,986    3,317        70,000
DEC    4000/610 AXP    (OSF/1)            2,198    3,247        70,000
DEC    4000/610 AXP                 2     3,816    6,215
 
DEC    7000/610 AXP                       2,189    4,217       150,000
DEC    7000/610 AXP    (OSF/1)            2,572    4,179       150,000
DEC    7000/610 AXP                 2     4,291    8,135      >150,000
DEC    7000/610 AXP                 3     6,307   11,860      >150,000
DEC    7000/610 AXP                 4     8,369   15,739      >150,000
 
DEC    10000/610 AXP   (OSF/1)            2,765    4,368       300,000
DEC    10000/610 AXP                4     9,108   17,187      >300,000
 
 
HP     9000/715/50                          866    1,710        15,000
HP     9000/725                             866    1,710        20,000
HP     9000/735                           1,832    2,950        41,000
 
HP     887S                               1,854    3,490
HP     890                          1     1,215    1,180
HP     890/2                        2     2,253    2,360
HP     890/4                        4     4,301    4,685
 
 
IBM    RS/6000 355                          961    1,936        20,000
IBM    RS/6000 365                        1,148    2,301        23,000
IBM    RS/6000 375                        1,332    2,612        27,500
IBM    RS/6000 970B                       1,117    2,220        94,000
IBM    RS/6000 980B                       1,404    2,960       114,000
 
 
SGI    Crimson R4000                      1,383    1,459        29,000
 
SGI    Challenge L (R4000)          4     5,562    6,131      <110,000
SGI    Challenge L (R4000)          8    10,177   12,020      <190,000
SGI    Challenge L (R4000)         12    13,406   17,370      <270,000
 
 
DG     Aviion  532                  2       933      621
DG     Aviion 5225                  2       868      532
DG     Aviion 5240                  4     1,591      971
DG     Aviion 6280                  8     3,245    -----
 
 
SNI    MX500-90/2                  12     4,870    -----
SNI    MX500-90/2                   4     1,919
 
*****************************************************************************
  end of table
*****************************************************************************
 
So with price no object, the winners in SPECrate_int are:
 
SGI    Challenge L (R4000)         12    13,406   17,370      <270,000
SGI    Challenge L (R4000)          8    10,177   12,020      <190,000
DEC    10000/610 AXP                4     9,108   17,187      >300,000
DEC    7000/610 AXP                 4     8,369   15,739      >150,000
Sun    SparcCenter 2000             8     8,047   10,600       197,000
DEC    7000/610 AXP                 3     6,307   11,860      >150,000
SGI    Challenge L (R4000)          4     5,562    6,131      <110,000
Sun    SparcStation 10/54           4     5,063    6,142        46,000
 
Price no object, SPECrate_fp:
 
SGI    Challenge L (R4000)         12    13,406   17,370      <270,000
DEC    10000/610 AXP                4     9,108   17,187      >300,000
DEC    7000/610 AXP                 4     8,369   15,739      >150,000
SGI    Challenge L (R4000)          8    10,177   12,020      <190,000
DEC    7000/610 AXP                 3     6,307   11,860      >150,000
Sun    SparcCenter 2000             8     8,047   10,600       197,000
DEC    7000/610 AXP                 2     4,291    8,135      >150,000
DEC    4000/610 AXP                 2     3,816    6,215       >70,000
Sun    SparcStation 10/54           4     5,063    6,142        46,000
SGI    Challenge L (R4000)          4     5,562    6,131      <110,000
Sun    SparcCenter 2000             4     4,539    5,627       130,000
 
Price under $50k, SPECrate_int:
 
Sun    SparcStation 10/54           4     5,063    6,142        46,000
Sun    SparcStation 10/512MP        2     2,950    3,744        34,000
Sun    SparcStation 10/412MP        2     2,411    2,854
Sun    SparcStation 10/402MP        2     2,112    2,378        25,000
DEC    3000/500 AXP    (OSF/1)            1,997    3,023        39,000
Sun    SparcStation 10/51                 1,847    1,930        28,000
HP     9000/735                           1,832    2,950        41,000
DEC    3000/400 AXP                       1,549    2,631        16,000
DEC    3000/300 AXP    (OSF/1)            1,535    2,137        10,700
DEC    3000/400 AXP    (OSF/1)            1,763    2,662        16,000
DEC    3000/500 AXP                       1,762    2,967        39,000
 
Price under $50k, SPECrate_fp:
 
Sun    SparcStation 10/54           4     5,063    6,142        46,000
Sun    SparcStation 10/512MP        2     2,950    3,744        34,000
DEC    3000/500 AXP    (OSF/1)            1,997    3,023        39,000
DEC    3000/500 AXP                       1,762    2,967        39,000
HP     9000/735                           1,832    2,950        41,000
DEC    3000/400 AXP    (OSF/1)            1,763    2,662        16,000
DEC    3000/400 AXP                       1,549    2,631        16,000
IBM    RS/6000 375                        1,332    2,612        27,500
 
============================================================================
 
        From: mas at everest.esd.sgi.com (Michael Schulman)
        Date: Wed, 5 May 93 11:00:20 -0700
        Subject: Re: Indigo vs. SPARC (Novice's question)
 
              IRIS Indigo R4K-XZ             SS10-41 GX
 
mips                85                        96
mflops              16                        17
specint92           58                        53
specfp92            61                        65
 
YOU might think that the sun is faster, but ...
 
load 3D model       53 secs                67 secs
recreate part      195 secs               260 secs
rotate              12 secs                77 secs
(20 frames)        (shaded)               (wirerame)
 
Micha



More information about the Bio-soft mailing list