AMD opteron with piledriver micro arch is a poor choice for consistent performance. It’s a refinement of bulldozer, which introduced a strange distinction between cores and modules. There’s a lot of sharing involved. (There’s even a lawsuit that what AMD sold as 8 core CPUs are really 4 core/8 thread CPUs.)
You’re having trouble linking the concepts of “performance” and “benchmark”? Besides the sqlite benchmark that is I/O bound, the architecture you’re shitting on shows up consistently among the top performers.
That works to a degree, but you still have to service interrupts and i/o. So generally have to knock a cpu out to dedicate to those things. Then if you have a NUMA system you get to deal with that too.
AMD opteron with piledriver micro arch is a poor choice for consistent performance. It’s a refinement of bulldozer, which introduced a strange distinction between cores and modules. There’s a lot of sharing involved. (There’s even a lawsuit that what AMD sold as 8 core CPUs are really 4 core/8 thread CPUs.)
Not at all. Take a look at my overclocked FX-8320E@4.4GHz - http://openbenchmarking.org/result/1601020-HA-1512297GA19&obr_sor=y&obr_hgv=FX-8320E%404.4GHz
And you can bet that those 3 processes showing performance issues were scheduled on different modules, so that’s not it.
I’m afraid I have no idea what I’m supposed to be looking at.
You’re having trouble linking the concepts of “performance” and “benchmark”? Besides the sqlite benchmark that is I/O bound, the architecture you’re shitting on shows up consistently among the top performers.
What about setting the affinity of each process to a certain core so the CPU scheduler is no longer intruding?
https://en.wikipedia.org/wiki/Processor_affinity
That works to a degree, but you still have to service interrupts and i/o. So generally have to knock a cpu out to dedicate to those things. Then if you have a NUMA system you get to deal with that too.