Mill Computing, Inc. › Forums › The Mill › Architecture › Pipelining › Reply To: Pipelining
Ivan,
Thanks for clarifying. “Effectively unbounded ILP in the loops that matter” may be more accurate, but admittedly doesn’t have the punch that “unbounded ILP” has. IMHO, Intel’s increase to eight micro-ops peak per cycle (e.g. in Haswell) shows that they think there’s enough detectable parallelism in x86 code to warrant the increased power and chip area required.