Benchmarks

In order to explore whether use of integer or floating-point code makes much difference to performance, I have done some crude Java benchmarks, of integer and floating-point performance on my two machines.

The results are as follows:

Test	AMD int	AMD float	G4 int	G4 float
Addition	40	250	1870	2006
Scalar multiplication	101	310	1891	1973
Squaring	130	361	1911	1911
Square root	711	931	2903	8369
Scalar division	40	511	2298	2319
Division	671	250	2304	2075
Shift right	50	511	1914	2318

AMD: AMDK6II, 400Mhz, 100Mhz memory, with the Microsoft JVM.

G4 = "whatever's in the Mac", with the built-in JVM.

My benchmarks are terribly unsophisticated - but should give the general idea.

For details you can look at the Java source code.

Summary

Integer performance is better on the x86. For some reason, floats beat integers on the division test, but not on anything else.

On the G4, floating point performance is much more nearly comparable to integers, but use of floats is still uniformly slower - excepting this time division (where floats come out slightly ahead) and squaring (which proved to be a dead heat).

According to these crude tests, Java performance on the G4 sucks - despite its faster and better processor.

Index | Links

tt@cryogen.com | http://alife.co.uk/