Very Computer

Quote:> I have just got a copy of the September 1990 SPECwatch and I am
> a bit concerned about the following paragraph:

> "There also seems to be a problem with replicating IBM's RS6000
> SPECmark results, and with achieving the expected levels of
> performance with other code.

While I haven't had a chance today to run the entire SPEC 1.0 suite on my
RS/6000-520 (GA code), I did run the following benchmarks from the suite.

                008.espresso
                013.spice2g6
                023.eqntott
                030.matrix300
                047.tomcatv

The SPEC ratios I got differ from IBM's by no more than 0.4.  I got the
same result for one test; one of my results was higher by 0.3, another by
0.4; and two were lower by 0.3.  (I think SPEC reporting rules keep me
from publishing an incomplete report.)

Vic Abell
Assistant Director
Purdue University Computing Center
SPEC license 310

Quote:> I have just got a copy of the September 1990 SPECwatch and I am
> a bit concerned about the following paragraph:

> "There also seems to be a problem with replicating IBM's RS6000
> SPECmark results,

I have been able to reproduce the SPEC ratio for the RS/6000-520 with
only minor changes needed to enable gcc.  (See the Notes/Summary of
Changes section in the report for gcc details.)

Here is a complete SPEC report for the RISC System/6000, model 520
POWERserver.  Its SPEC ratio is 0.1 larger than what IBM reported to me.

                    SPEC Throughput Method A# Results
                        for Release 1.0 Benchmarks

Results:       SPEC    IBM RS/6000                             IBM Corporation
               Ref.        520    
Benchmark      Time    Time    SPEC                       IBM RISC System/6000
No. & Name    (sec.)  (sec.)  Ratio                            POWERserver 520

001.gcc        1482     102    14.5*              Hardware
008.espresso   2266     139    16.3   Model Number:      POWERserver 520
013.spice2g6  23951    1189    20.1   CPU:               IBM POWER 20MHz
015.doduc      1863      88    21.2   FPU:               Integrated
020.nasa7     20093     771    26.1   Number of CPU's:   1
022.li         6206     398    15.6   Cache Size/CPU:    32 KB data, 8 KB ins.
023.eqntott    1101      60    18.4   Memory:            32 MB
030.matrix300  4525     263    17.2   Disk Subsystem:    2-857 MB SCSI
042.fpppp      3038      71    42.8   Network Interface: 1 Ethernet Controller
047.tomcatv    2649      47    56.4

Geometric Mean 3867.7   173.0  22.4*

            Software                                System
O/S Type and Rev:  AIX 3.1, GA        Tuning Parameters: None in use
Compiler Rev:      XL Fortran 1.1     Background Load:   Normal Unix daemons
Other Software:    XL C 1.1           System State:      Multi-user, lightly
File System Type:  IBM Journaled                         loaded
                   File System
Firmware Level:    N/A

Tested in:      October 1990

                Purdue University Computing Center
                Mathematical Sciences Building
                West Lafayette, IN 47907
                (317) 494-1787
SPEC License #  310

Notes/Summary of Changes:

  # Method A: Homogeneous Load

  * Portability changes were required:

    gcc:
        o  Modified alloca.o rule in Makefile to use XL C compiler.
        o  Activated #include of <time.h> in cccp.c.
        o  Disabled redefinition of ptrdiff_t and size_t in stddef.h.
        o  Modified erroneous assignment of enumerated variables to bit
           fields in rtl.h, tree.h, and varasm.c.
        o  Enabled vfork to fork redefinition in gcc.c.

Copyright 1990 Purdue Research Foundation, West Lafayette, Indiana 47907.
All rights reserved.