Post BAFbt7raGcvjLNYLXk by natty@astolfo.social
 (DIR) More posts by natty@astolfo.social
 (DIR) Post #BAFbt7raGcvjLNYLXk by natty@astolfo.social
       0 likes, 0 repeats
       
       @solonovamax@tech.lgbt @theking@mskey.nekomimi.party amdahl's lawIf you have a very parallelized workload, it's still bottlenecked by the single threaded bitsAnd the core layout follows the heuristic that usually you do one or two things at a time, so the (by definition) synchronous tasks that you interact with get pinned to the fast coresAnd if you do have parallel workloads, it is much less power to feed the atom coresThe fast cores aren't even all that useful for hyperthreading if you can (or rather, can't, since HT/SMT assumes the workload doesn't choke on the shared pipeline bits) split the workload into parallel pipelines to begin withIt also saves silicon space because the perf core SMT proceeds to lose the perf part to shit like spectre