You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I looked into the code a bit and it seems that nvp or num_nvp to be exact, is used in terms of "total number of virtual processes per node". In plots, until now we were mainly interested in the scaling behavior of the number of virtual processes per node while keeping the number of nodes the same. This is also why in bennch for example, there are two scaling modes, one called "threads" the other called "nodes". It might still be interesting though to see the scaling behavior of both threads and nodes together, i.e., the actual total number of virtual processes.
Maybe we should add another column called local_num_nvp or something similar and make num_nvp the actual total number as you suggested Jari.
The number of virtual processes nvp is calculated like this:
This is the number of processes per node, but shouldn't it rather be the number of processes across all nodes?
The text was updated successfully, but these errors were encountered: