[david-gpu]

> There should be an O notation which takes into account everything - various kinds of memory latencies, speed of logic operations

O notation is useful as-is because it simplifies an algorithm down to the bare bones. If such simplification doesn't capture what is important to you, you need a more complex model.

The folks designing GPUs use multiple different such models, ranging from the simple, inaccurate, understandable and fast; to the obnoxiously detailed, precise, abstruse and slow.

Not one model will ever have all the desirable properties, because they are in direct contradiction.