It isn't clear if Haswell has this optimization. As an easy test you could run the zero-fill-bench linked in the post and check the fill0 vs fill1 numbers. If t

It isn't clear if Haswell has this optimization. As an easy test you could run the zero-fill-bench linked in the post and check the fill0 vs fill1 numbers. If they are the same, Haswell probably never had this optimization in the first place. Based on the Intel microcode release note [1] it doesn't seem like client HSW got any update this time, only HSX (Haswell Xeon). --- [1] https://github.com/intel/Intel-Linux-Processor-Microcode-Data-Files/commit/3f97690f0da8011f52209b232450a1e5c4f2e1f6#diff-e6fad0f2abbac6c9603b2e8f88fe1d151a83de708aeca1c1d93d881c958ecba4 https://github.com/intel/Intel-Linux-Processor-Microcode-Dat...