Whetstone Benchmark Java Version
For the benchmark to run, Java Runtime Environment (JRE) or Java Virtual Machine (JVM) needs to be installed. These and other security settings might also need to be enabled. The results will be on a blue background below. To save them, use print/copy screen facilities and paste into a graphics package.
To rerun, click on the Refresh/Reload Current Page icon. WARNING - There can be a considerable delay before the benchmark actually starts.



This is a revised version of the on-line benchmark produced in 1997. See below for details. Old and new benchmarks, including the Java source code, can be found in:

onlinetests.zip for Windows or onlinetests.tar.gz for Linux

Introduced in 1972, the original Whetstone Benchmark provided a performance rating in terms of Million Whetstone Instructions Per Second or MWIPS. This, and other later versions, also measure the speeds of individual test functions in Million Operations or Floating point Operations Per Second (MOPS or MFLOPS). Note that, using the same software and processor model, measured speeds can be expected to be proportional to CPU MHz. See the following:

Whetstone Benchmark History and (Old) Results

Whetstone Benchmark Detailed Results On PCs - Java

Java Performance - Comparison with other programming languages

Roy Longbottom’s PC Benchmark Collection

Output Format

The benchmark was also compiled to display results in a Windowe Command Prompt or Linux Terminal display, via a “java whetstc” command, with details also being saved in a whets.txt log file. All outputs have the same format, as follows:

     Whetstone Benchmark Java Version, Dec 8 2011, 15:39:00

                                                       1 Pass
  Test                  Result       MFLOPS     MOPS  millisecs

  N1 floating point  -1.124750137    905.66             0.0212
  N2 floating point  -1.131330490    720.26             0.1866
  N3 if then else     1.000000000            1005.83    0.1029
  N4 fixed point     12.000000000            1923.08    0.1638
  N5 sin,cos etc.     0.499110132              43.72    1.9030
  N6 floating point   0.999999821    500.84             1.0770
  N7 assignments      3.000000000             643.90    0.2870
  N8 exp,sqrt etc.    0.825148463              33.57    1.1080

  MWIPS                             2062.07             4.8495

  Operating System    Windows 7, Arch. amd64, Version 6.1
  Java Vendor         Sun Microsystems Inc., Version  1.6.0_21
  CPU AMD64 Family 16 Model 4 Stepping 2, AuthenticAMD

  Or

  Operating System    Linux, Arch. amd64, Version 2.6.35-24-generic
  Java Vendor         Sun Microsystems Inc., Version  1.6.0_20
  CPU null
         
The numeric results are derived from a constant minimum number of passes and could be expected to be consistent on all variations of hardware and browsers. One exception was found on a Core 2 Duo based laptop, just running Ubuntu 10.10, where the N8 result was 0.935364604. Tests other than those produced 0.825148463.

Old and New Version Differences

The original Java benchmark was run for 100 seconds, needed because of the wide variation in the time taken by the different tests and possible timer resolution issues. There was a Run button to start testing. It also provided tests to measure graphics speed. When run under Linux, using the Firefox browser, there were text formatting issues and false speed indications due to compiler over-optimisation.

In the latest version, slight changes were made to the programming code to avoid the optimisation issues. This time, all tests are run for a minimum of one second, with running time milliseconds being adjusted to determine speeds of a single pass, using the original loop count parameters. Execution of the benchmark starts automatically on loading, sometimes with an inexplicable delay, or after clicking on the refresh symbol.

NOTE - Using both on-line versions, displayed results might be obliterated if the page is resized, moved, scrolled or overlaid by another window.



Roy Longbottom December 2011