When a numerical program is run in multiple environments (operating systems, architectures, or compilers), assessing its consistency can be a difficult task for a human, since small differences in numerical output values are expected.

Application of a file differencing utility, such as POSIX/UNIX diff(1), will generally produce voluminous output, often longer than the original files, and is thus not useful. The lesser-known UNIX spiff(1) utility, while capable of handling numeric fields, suffers from excessively-long running times, and often terminates prematurely.

ndiff provides a solution to this problem. It compares two files that are expected to be identical, or at least, numerically similar. It assumes that lines consist of whitespace-separated fields of numeric and non-numeric data.