2 ms·
Is their LU benchmark just LU factorize and possibly solve? That seems like an odd candidate for this kind of problem, because of course anybody running that ki
by bee_rider 3y ago
Is their LU benchmark just LU factorize and possibly solve? That seems like an odd candidate for this kind of problem, because of course anybody running that kind of problem that cares about performance just doesn’t over-subscribe their system… the scheduler has a much easier job when there’s one big chunky thread per core.
I mean, I get that they are testing a general fix for the sort of problems that, like, non-scientific-computing users get. So it isn’t to say the work is useless. It just seems like a funny example.
I think coming up with benchmarks for schedulers must actually be very difficult, because you have figure out exactly how silly of a thing you want the imagined user to do.
- semi-extrinsic 3y agoFurther to this, if you are running a program where compute is mainly a sparse linear solve, you are memory bandwidth limited and on modern systems like AMD Epyc you will very very likely see improved performance if you undersubscribe cores by 2x - e.g. using max 64 threads per 128 cores.