Document:         BENCHDOS.ZIP
File Group:       Classic Benchmarks
Creation Date:    3 February 1997
Revision Date:    11 March 1997

Title:            Classic Numeric Benchmarks, Compiled for DOS
Keywords:         BENCHMARK PERFORMANCE LIVERMORE WHETSTONE DHRYSTONE
                  LINPACK MIPS MFLOPS

Abstract:         This suite of Classic Benchmarks has been compiled to
                  operate under DOS.  The suite includes Whetstone,
                  Dhrystone, Livermore Kernels, and Linpack.  For
                  information on these benchmarks refer to the file
                  BENCHSCI.TXT in this library.

Note:             This suite is still under test and should be treated as
                  beta test software.  Please submit feedback and data sets
                  directly to Roy, or as a posting in Section 12.

                  Results TXT files are inserted in the directory from
                  which the benchmarks are run. These files contain numeric
                  results produced by the benchmark and details of the run
                  time environment. The latter has to be typed in each time
                  a benchmark is run. As this is rather tiresome, when
                  submitting a number of results, it is only necessary to
                  provide complete details once. In this case, it would be
                  useful if runs without full details have some cross
                  reference. Minimum requirements for the "complete" data
                  entered are CPU chip type and MHz. An abbreviation of
                  these is suitable for a cross reference in other tests.                              
                  You may use other entry fields for your own reference
                  purposes.

                  The latest versions of the benchmarks can also be executed
                  with a run time parameter N to omit the typing in phase.
                  A batch file BATCHxxx.BAT is supplied to execute all
                  benchmarks. This also runs program SYSTEMxx.EXE for
                  entering configuration details once for the whole batch,
                  the results being saved in BENCHSYS.TXT.
                  
WARNING:          The benchmarks have been tested on various PCs including
                  386, 486, Pentium and Pentium Pro systems. They have also
                  been run via Windows 3.1, Windows 95,  NT 3.51 and 4.0.
                  The DOS versions have also been run directly in the
                  associated DOS mode and when booted as DOS only with MS
                  DOS 5.0 and 6.2.

                  However, it is inevitable that they will not run properly
                  in all circumstances or occasionally may cause certain
                  systems to hang. The most likely reasons for failure are
                  hardware or software incompatibilities, particularly
                  those associated with memory management. Remember all the
                  problems that you have has with commercially available
                  software or games that are supposed to have been
                  thoroughly tested. The latter tend to have long lists of
                  incompatibilities.

                  The DOS versions of the benchmarks are the most
                  vulnerable to failure as the Windows mechanisms for
                  avoiding memory conflicts are missing and DOS only PCs
                  are likely to make use of all kinds of old software
                  routines. For these, the same rules apply that are
                  normally associated with games - produce a clean DOS boot
                  disk, free from all drivers and memory management
                  systems. If necessary, disk compression software can be
                  excluded and programs run from copies on a floppy disk.

                  It should also be remembered that, in running all the
                  benchmarks, the CPU may never have had such an exhaustive
                  test, so the possibility of strain related hardware
                  faults cannot be ignored.

                  In the event of faults being encountered, it is important
                  to know about which benchmarks ran properly besides the
                  ones that failed (and to see the successful results
                  files). Both optimized and non-optimized versions of the
                  benchmarks should also be tried as they may be compiled
                  with different memory models. The following running order
                  is suggested:

                  All Dhrystones - these do not use much memory and have
                                   no floating point instructions.

                  All Whetstones - these do not use much memory but require
                                   the math coprocessor for floating point.

			All Linpack    - these use much more memory for data and
                                   are mainly floating point calculations.

			All Livermore  - these are the largest users of memory,
                                   including 600 KB for data. They also are
                                   mainly floating point and have, by far,
                                   the longest running time.


Contributor:      Roy Longbottom 101323.2241@compuserve.com
                  Roy_Longbottom@compuserve.com


