[Beowulf] Broadwell HPL performance

[Mark Hahn]

> cores can stabilize turbo frequencies above nominal clock. The turbo
> boost is more pronounced the less cores you have (less heat).

but is it based on heat or on some internal measure of (instruction) load?
from Intel docs (albeit in earlier generations), the narrative seems to be 
that turbo is disabled if you use more than a smallish fraction of the cores.
ie, that you get no turbo at all unless you have several fully-idle cores.
(not sure whether they have to be in HALT, or just a lower power state.)

it would be easier to manage if temperature-based (and superior cooling 
would pay off in more time-in-turbo.  I'm not clear on whether the
AVX-downclocking thing also might me temperature based.  (or perhaps power
based?)

> Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
> 90000  90000  4      699.126    695.1769 6.759477e-09 2.983269e-02   pass

nice!  and thanks.