units branch integer floating point on 601 issue at most one per unit per cycle eight entry instruction queue can fill queue from cache in one clock cycle loads from requested address to end of cache block pipeline prefetch includes ins. cache access cycles decode execute writeback fpu IQ[3210] → fpu buffer/decode [≥1 cycle] → execute 1 → execute 2 → writeback iu IQ0/decode → buffer [if exec busy] → execute [hold for dependency] → circulate in load/store writeback bpu IQ[3210] → decode/execute → writeback notes address calculation must complete before stored value enters write buffer