Factor Language Blog

I ran a number of Factor benchmarks on the following machines:

• 2.8 GHz Pentium 4, running Debian Linux, proprietary NVidia driver
• 2.4 GHz Athlon 64 (dual processor), running Ubuntu Linux, proprietary NVidia driver
• Apple MacBook Pro - 2.0 GHz Intel Core Duo (dual processor) running Mac OS X
• Apple PowerMac - 2.5 GHz G5 (quad processor) running Mac OS X

The benchmarks were:

• Stage 2 bootstrap - this involves compiling all words, building the online help search index, and loading the UI backend.
• Fibonacci - this tests procedure activation overhead and integer math.
• Mandelbrot - this tests complex number arithmetic.
• Raytracer - this tests floating point arithmetic.
• Evaluating [ tests compiler-tests ] time in the UI listener - the unit tests themselves include a lot of performance-intensive code, and this also measures how fast the UI can relayout and redraw, since the tests generate a lot of output.

The results were interesting. The best result in each row is shown in italics.

	Pentium 4	AMD64	MacBook Pro	PowerMac G5
Bootstrap (mins:secs)	4:56	4:40	4:38	6:19
Fibonacci (ms)	360	320	470	830
Mandelbrot (secs)	3.3	2.8	2.4	5.4
Raytracer (secs)	18	21	18	42
Unit tests (secs)	24	22	20	38

The results indicate there is a lot more work to do on the PowerPC compiler backend. The Core Duo’s good performance is rather surprising.