[Beowulf] Parallel programming for Xeon Phis

Douglas Eadline deadline at eadline.org
Wed Aug 24 09:32:28 PDT 2016


Understanding your code and optimization for HW is where
future progress in HPC is going to come from. Throwing
more and more hardware at a problem has a diminishing return.

I recent wrote a paper for InsideHPC/Mellanox about the
co-design process where machine designers, coders and application
specialist are literally in the same room working on new
systems (Summit and Sierra)

Back in the day (okay I said it), when there were just a
handful of supercomputer companies, each company would
have whole teams that would optimize potential customer codes
for their machines. On many customer codes they would get
huge speedups (the top 20 known packages were well understood
on the target machine).

I was convinced at one point that customers just roughed out
the code, and let the companies optimize for them. But then
the customer was plunking down 6-7 figures for a system
so code optimization was probably the cheap part of
the sales effort.

Okay, now get off my lawn.

--
Doug


>
> his is an old article, but it's relevant to the recent discussion on
> programming for Xeon Phis, 'code modernization', and the speedups 'code
> modernization' can provide.
>
> https://www.hpcwire.com/2015/08/24/cosmos-team-achieves-100x-speedup-on-cosmology-code/
>
> --
> Prentice
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf
>
> --
> Mailscanner: Clean
>


--
Doug

-- 
Mailscanner: Clean



More information about the Beowulf mailing list