These are the Whetstone, Dhrystone, Linpack and Livermore Loops benchmarks
(see
Classic.htm and
Whetstone.htm
for details plus BenchNT.zip for
latest versions that run via Windows). Further details and results can be found
via
Main Page. The following versions are available
with optimised and non-optimised varieties:
In DOSTests.zip Classic Benchmarks compiled for DOS
In OS2Tests.zip Classic Benchmarks compiled for OS/2
Classic Benchmarks compiled to use 16 bit instructions
Dhry1C16 Dhrystone 1 via Watcom 10.5 C/C++
WhetC16.exe Whetstone via Watcom 10.5 C/C++
WhetBAS.exe Whetstone BASIC via BASCOM
WhetProf.exe Whetstone FORTRAN via PROFORT
WhetMS5F.exe Whetstone FORTRAN via MS 5.1
A summary of results, including those for OS/2 and 16 bit versions, are in the following
files. DOS and Windows version results are not identified as performance is little different.
The source codes for the Whetstone, Dhrystone, Linpack and Livermore Loops Classic Benchmarks
are included in BenchNT.zip
see Main Page.
Documentation in benchmark zip files also include examples of the code used.
WhetJava.zip, Whetstone Benchmark Java version contains:
Whets.htm (HTML description and running facility),
Whets.cla (CLASS benchmark code) and
Whets.jav (JAVA Whetstone benchmark source code).
WhetsVB.xls is an Excel Spreadsheet containing the Whetstone Benchmark in Visual Basic.
WhetsVB.txt contains the source code for use in standard Visual Basic. Whets.c
Other Whetstone Benchmark source code in files Whets.for and Whetd.for are single and double precision Fortran,
then Whets.bas is in Basic. Whets.c is a C/C++ variety with timers other than those for PCs.
BusSpeed measures memory and cache bus data transfer rates.
Superseded by BusSpd2K (
see Main Page) which includes the same assembly code procedures and a greater variety of memory demands.
BusDOS.exe, compiled for DOS, is now in DOSTests.zip and BusOS2.exe, compiled for OS/2, in OS2Tests.zip.
The DOS version can be used as a substitute for the Windows 9X/NT version. The three versions were previously in BusSpeed.zip.
MemSpeed measures memory and cache speeds with moderate processing.
The file contains description and benchmarks which
essentially use the same code as MemSpd2K (
see Main Page) but only tests up to 2 MB.
Memory for the two arrays is not aligned on MB boundaries so performance differences can be apparent.
MdtrDOS.exe, compiled for DOS, is now in DOSTests.zip and MdtrOS2.exe, compiled for OS/2, in OS2Tests.zip.
The DOS version can be used as a substitute for the Windows 9X/NT version. The three versions were previously in MemSpeed.zip.
MMXSpeed measures CPU cache and memory speeds using all MMX instructions.
MMXMax measures maximum memory speed and is superseded by
BusSpd2K (
see Main Page). Example source code is included in the documentation. Contents are:
MMXDOS.exe and MMXMax, compiled for DOS, is now in DOSTests.zip and MMXOS2.exe, compiled for OS/2, in OS2Tests.zip.
The DOS version can be used as a substitute for the Windows 9X/NT version. The three versions were previously in MMXSpeed.zip.
Disk98.zip for Win 9X/NT/2K/XP
Two versions are available DiskNT98.exe and DKFat16.exe,
the only difference being a function that determines free disk space.
The benchmark can be run from menu selections or a command line
to measure serial transfer rates and random access times of varying
size files. Superseded by DiskGraf (
see Main Page)
but has many parameters useful for experimentation.
DiskRand DiskMBPS
These are older disk benchmarks. The DOS versions indicate very low speeds on modern disks unless a DOS based DMA bus driver is available. They may be of use in checking disks before Windows is loaded but cannot be used where the disk is formatted for NTFS.
DiskMBPS writes a single file and reads it multiple times. Small files demonstrate Windows File Cache memory speed.
The benchmarks are DiskDOS.exe (compiled for DOS) and DiskOS2.exe (compiled for OS/2).
DiskDOS can be use instead of DiskNT95.exe, compiled for Windows 9X/NT. The files were
originally in DiskMBPS.zip. They are now in DOSTests.zip and OS2Tests.zip.
DiskRand measures disk random access times. It also has facilities to emulate transactions accessing a database via multiple users.
Tests are RandDOS.exe (compiled for DOS) and RandOS2.exe (compiled for OS/2. The former can be used in place of RandNT95.exe,
compiled for Windows 9X/NT). These are also in DOSTests.zip and OS2Tests.zip instead of DiskRand.zip.
DOSTests.zip contains the following VideoDOS benchmark besides descriptions and
results. These were previously in VideoDOS.zip
1 - VideoDOS.exe - DOS function tests - measures speed in terms
of Millions of Pixels Per Second for putimage, ellipse,
rectangle (fill), setpixel, lineto and getimage/putimage at 16 and 256
bit colour settings.
See Windows Video Tests in
Main Page where VideoWin executes similar functions via Windows.
VDUTest.zip Monitor Quality Test and 16 Bit Benchmark
Contains VDUTest.exe, a monitor quality test that displays bitmap patterns with full screen tiling and zoom in and out. BMP pattern files are included.
The zip file also includes
VideoW16.exe a 16 bit version (for Win 3.1 and later) of VideoWin
See Main Page.
BurnDOS and BurnOS2 are reliability tests compiled for DOS and OS/2.
They contain floating point and disk burn-in test programs that check data
for correctness and can be run for a long time. The test functions are
the same as included in Burn9XNT, the revised version for
Windows (
see Main Page).
BusSpeed benchmark. with CPU, Cache and RAM tests
include some where data is checked for correctness via DOS and OS/2.
BurnDOS is now in DOSTests.zip and BurnOS2 in OS2Tests.zip rather than
BurnDOS.zip and BurnOS2.zip. These ZIP files also include BusSpeed tests,
Results are available for PCs using 80386 or 80486 CPUs up to modern
processors.
See Main Page.
Some of the files include older benchmark results. The results files are:
Three sets of data are provided for mainframes, minicomputers, supercomputers, workstations and PCs. Where available, production year is also shown. Speeds are in Kilo (thousands) of Instructions Per Second (KIPS), Millions of Instructions Per Second (MIPS) or Millions of Floating point Operations Per Second (MFLOPS). They are:
CPUMix - Processor speed calculations in KIPS, based on mixes of instructions for commercial and scientific applications for some computers from pre-1960 to the 1970's.
MIPS - Published speed claims for more than 2000 computers produced between 1980 to 1996. Costs and other configuration details are also shown.
Whets - Whetstone benchmark results mainly 1970's to 1990's. From 1980's results include derived MIPS and MFLOPS ratings. An extra section has been added just for PCs, to include results up to 2005.
These are in HTML format with indexed look-up facilities:
Details of IBM痴 larger mainframes and PC CPUs up to 2004 have also now been included in the latter.
Calculations for sizing or capacity planning of on-line computer systems invariably involve the use of a queuing model in order to estimate response times and throughput. In most cases, readily available formulae from standard queuing theory are used. The most frequently used calculation is to estimate the increase in service time as 1/(1-u), where u is device (e.g. CPU) utilisation. Unfortunately, the queuing theory is based on the assumption that there is an infinite number of customers, the calculation producing infinite queuing time with a utilisation of 1.0 (100%). There are formulae for finite populations but these are more complicated. Even more complexity is introduced when considering multiple processors.
During the 1980s, Roy Longbottom of CCTA turned existing models inside out and introduced "Think time" to enable calculations to be carried out for transaction processing and RDBMS applications. In the case where each stream is considered to be executing the average transaction profile, the calculations are quite simple. The models were used on numerous occasions for sizing new computer systems, validating on-line and batch benchmarks and in capacity planning.
The spreadsheet model has input parameters for think time, CPU time, disk time etc. for the average task or transaction. It carries out calculations for an increasing number of tasks or on-line users. These are queuing times, response time, elapsed time, device utilisation and queue lengths. The model is intended for use as an aid for sizing and capacity planning of on-line computer systems, also in understanding multitasking benchmarks. XLS and WKS spreadsheet versions are supplied.