On Fri, Mar 08, 2013 at 11:44:12AM -0500, Ted Unangst wrote: > On Fri, Mar 08, 2013 at 11:54, Tobias Ulmer wrote: > > > I don't believe it. Someone please compile a kernel with > > time make -j$(sysctl -n hw.ncpu) with or and without HT and report > > back... (and please, not on a space heater xeon 4 from 2003...) > > If you are splitting a core, your process will run about 50% slower. > (On Atom, anyway.) But you will be running more processes at once. > > On a machine with a single core, this is strictly a win. e.g., > Compiling a file takes 1 second. Compiling two at the same time with > HT takes 1.5 seconds. Compiling two sequentially without HT takes 2 > seconds. 1.5 < 2. qed. :) > > If you have two or more cores, you have to balance this fact with your > ability to load all cores. >
Did a little benchmarking: hw.model=Intel(R) Core(TM) i5 CPU M 520 @ 2.40GHz No HT, -j2: 2m2.25s real 3m14.05s user 0m23.09s system 2m2.58s real 3m15.89s user 0m22.23s system 2m4.67s real 3m18.11s user 0m23.43s system No HT, -j1: 3m46.46s real 2m38.15s user 0m18.60s system 3m46.46s real 2m39.36s user 0m18.81s system 3m46.33s real 2m38.07s user 0m19.22s system HT, -j1: 3m48.97s real 2m32.24s user 0m17.35s system 3m47.57s real 2m32.95s user 0m15.58s system 3m47.49s real 2m32.77s user 0m15.82s system HT, -j4 1m43.51s real 5m20.49s user 0m47.84s system 1m45.53s real 5m26.49s user 0m47.84s system 1m47.28s real 5m31.17s user 0m48.16s system I know which setting I would use. zzz
