Of course, these numbers are all misleading - the above processors do floating point math in 64-bit. Also, modern consoles have extended instruction sets (or, in the case of the PS 2, extra chips) to provide vector math. But even vector math is generally not more than 32 bits of precision - it's just that more elements are processed at once (see MMX Optimization Guide for details on one vector processor instruction set). I guess the point of this is: counting bits is as foolish as counting Mhz or MIPS.

Update: ikeleib - It's 128 bit SIMD, which means up to 32 bits of precision, but 4 at a time. The following article explains more: http://arstechnica.com/reviews/1q00/playstation2/ee-1.html

So, yeah, what I said above about the PS2 was an over-simplification. It's really 8-bit (yeah, you can do 16 8-bit ops in a cycle, if you really want to - it'll let you), 16-bit (it's even got a 16-bit data bus, as well as some larger data busses), 24-bit (color, Z-buffer (iirc)), 32-bit, 64-bit (although I don't know if that's really 64-bit, or just 2x32 bit SIMD), and 128-bit (although there are probably only 5 or 6 128-bit instructions: load, store, xor, and, or) - all at the same time.