You could argue that float/int arithmetic is inherently imprecise. It is, however, fully deterministic.
Of course as you hint in your comment, at a lower level hardware is prone to failures and lack determinism: bugs in the cpu, ram corruption (ECC-ram is of course better), data corruption everywhere on storage/network/data links, etc. Ever heard a gamer bitching about Windows 7 instability? That's not Windows, but overclocking not quite under control and/or insufficient cooling...
And when all is under control a cosmic ray might still wreak your bits havoc.
It's deterministic at some level, but can be very difficult to predict when a compiler will decide to spill 80-bit float registers to 64-bit doubles in memory.