Simon 
Just thought it might interest someone... I've been optimising a 3D engine (written only for the CPU, since we don't have OpenGL) and made some timings of integer vs float performance. The (trivial) code is at http://0x0000ff.com/src/math.c, and it's compiled and installed using: prompt% arm-linux-gcc -O2 -o t math.c prompt% lftp -c "open 192.168.1.11 -u 6410,6410; put t -o t" Here's what I get: [root@6410 /sdcard]# ./t Division: Ints: 0.894195 (162725365) Floats: 1.084420 (146314368.000000) Ratio: 1.212733 Multiplication: Ints: 0.095186 (0) Floats: 0.247450 (0.000000) Ratio: 2.599647 Addition: Ints: 0.076059 (-2014260030) Floats: 0.247430 (48714878025728.000000) Ratio: 3.253132 Subtraction: Ints: 0.076480 (2014260034) Floats: 0.247376 (-48714878025728.000000) Ratio: 3.234519 The 'Ratio' line is the speed-up of using integer vs floating point (I was investigating creating a fixed-point system, doesn't really seem worth it for these figures, given all the extra bit-shifting involved). Simon