Cache Replacement Policies

In a set-associative cache, when all ways in a set are occupied and a new block must be loaded, which replacement policy do most modern L1 caches use and why?

Optimal (Belady's), because modern hardware can predict future accesses

FIFO, because it is simplest to implement in hardware

LRU or pseudo-LRU, because recently used lines are likely to be used again

Random, because it avoids pathological worst cases that LRU suffers from