open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi to all,
I run T2d on a 24-processors computer
I made some tests and noticed that running the max number of processors may not always be the faster solution.
Are there some recommandations about the relationship between a model size (eg number of nodes) and the optimal number of processors to run it ?
My concern is how to save time by running various simulations at the same time.

Hello Benoit,

This answer depends on your computer and the processors inside. You have to find the scalability (if you run a computation and if using twice the number of cores and CPU time is divided into 2, you are still linear for scalability ; in that case you can try to double the number of cores to see if CPU time is still divided into 2 or less).

Anyway, with the computer I use, 5,000 to 10,000 triangles per subdomain is a good target for me but depending on how you want to use the cores, this target may be a little bit less.
I would not recommend under 1,000 (more time to communicate between subdomains rather than computing) but you may try.

Hope this helps,

Chi-Tuan