Actually, quite a lot of TAOCP are out of date. TAOCP's discussion of floating point arithmetic, while correct, predates IEEE 754 and not that useful now. TAOCP's analysis of algorithms, while correct, ignores locality of reference as it predates importance of CPU caches and branch predictors. TAOCP spends inordinate amount of pages on external sorting with tapes; while external sorting is still important, most of Knuth's specifics are not. I can go on and on...
In fact, these things are cyclical. Mainframe->microprocessors->network computing servers. Floating-point emulation using integers on the CPU->External floating-point coprocessors->on-chip math coprocessors->external GPUs. Manipulating gigantic data sets on resources like Amazon S3 has many commonalities with classic tape storage (arguably more than it does with rotating disk storage -- retrieving a big object from S3 brings it in as a stream of data. Reading the whole glob and processing it as a stream is reasonably efficient but random access sucks. Sounds like a tape, no?). IEEE 754 isn't necessarily the last word in floating point arithmetic. I can go on and on... :-)
Sometimes accountants don't get much respect, but I have to admire anyone who can gobble the Internal Revenue Code every year and come up with procedures that (usually) don't get any of their clients fined or jailed and (ideally) don't cause them to go bankrupt.