Thank you for the quick answers.
On 19.02.2015 20:12, Riccardo (Jack) Lucchetti wrote:
On Thu, 19 Feb 2015, Artur T. wrote:
> Dear Allin and Jack,
> I'have observed some increased efficiency in the built-in multi-core
> support during the last few months. Just out of curiosity: has really
> anything changed in this direction or is it simply my subjective
> impression that the CPU usage has substantially increased running the
> same script on the same machine?
> For instance, on a modern i5 4-core machine almost 2 cores are
> automatically exploited for heavy-loop and heavy-matrix computations.
> On an older AMD 2-core machine the CPU usage is slightly lower but
> still the 2nd core is at least used up to 40% or so.
Have you been seeing a noticeable increaswe in performance? I'm asking
because the benefits from parallelisation are in most cases far from
being linear, and we ran extensive tests on what the best strategy
was; often (matrix products, for example) we settled on compromise
solutions that appeard to work "well" on the hw/sw combos we had at
our disposal, but of course the more real-life cases we hear, the
better we can gauge multicore usage.
I have to admit that my impression is more or less subjective as I
haven't really measured it. My impression mainly arises from monitoring
the CPU usage using the task manager. This revealed a much better CPU
usage on Windows as well as Linux.
However, at work we have a Windows server running a modern Xeon-type CPU
which has really impressive properties in terms of frequency and number
of cores. Surprisingly, the computation of a heavy-loop and heavy-matrix
script is really much lower in comparison to the i5 4-core laptop CPU --
let's say about 20% or so. But this may be due to the "compromise
solutions" you mentioned.