[Beowulf] likwid vs stream (after HPCG discussion)

Mikhail Kuzminsky kus at free.net
Mon Mar 21 09:46:31 UTC 2022

In message from Scott Atchley <e.scott.atchley at gmail.com> (Sun, 20 Mar
2022 14:52:10 -0400):
> On Sat, Mar 19, 2022 at 6:29 AM Mikhail Kuzminsky <kus at free.net> 
>> If so, it turns out that for the HPC user, stream gives a more
>> important estimate - the application is translated by the compiler
>> (they do not write in assembler - except for modules from 
>> libraries), and stream will give a real estimate of what will be
>> received in the application.
> When vendors advertise STREAM results, they compile the application 
> non-temporal loads and stores. This means that all memory accesses 
> the processor's caches. If your application of interest does a random 
> through memory and there is neither temporal or spatial locality, 
> using non-temporal loads and stores makes sense and STREAM 

STREAM is not initially oriented to random access to memory. In this
case, memory latencies are important, and it makes more sense to get a
bandwidth estimate in the mega-sweep

More information about the Beowulf mailing list