Why only enough threads to "fill" parallel machine? Why not 100 threads per core?
yuel1
to add to that, I'm confused as well with the statement "Best practice: create enough workers to 'fill' parallel machine, and no more". In the first few lectures, we saw that having more threads than actual cores could be useful for hiding memory latency as also mentioned in the second bullet point...
Why only enough threads to "fill" parallel machine? Why not 100 threads per core?
to add to that, I'm confused as well with the statement "Best practice: create enough workers to 'fill' parallel machine, and no more". In the first few lectures, we saw that having more threads than actual cores could be useful for hiding memory latency as also mentioned in the second bullet point...