14:04 < axboe> Glanzmann: 2xcpus for compiles is somewhat of a relic of the days...

14:04 < axboe> Glanzmann: 2xcpus for compiles is somewhat of a relic of the days where we had rotating storage, and hence a lot of iowait time (mostly during linking, but also just paging in data) where it then makes sense to run more jobs in parallel
14:05 < axboe> on decent storage these days, -jN where N is close to the number of CPUs will usually give you the best results, with N+small_bump might be a tad faster, but 2N is usually somewhat slower
14:05 < axboe> (just my experience, and this is all based on compiling the kernel, ymmv on other things obviously)