Coarse-grained parallelisation efficiency in RAxML-NG

[Kenta Renard]

Dear All,

I was wondering about efficiency losses for coarse-grained parallelisation as you increase the number of threads. If I want to infer 30 trees with the --threads 15 --workers 15, what would the efficiency loss be (even if it isn't much) if I ran using --threads 30 --workers 30.

Best wishes,

Kenta

[Oleksiy Kozlov]

Dear Kenta,

depending on your dataset and hardware, the could be efficiency loss or efficiency gain.

Please read:

https://github.com/amkozlov/raxml-ng/wiki/Parallelization

In general, you could either rely on automatic parallelization, or try multiple configurations
yourself and compare the runtime.

Best,
Oleksiy

> --
> You received this message because you are subscribed to the Google Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to
> raxml+un...@googlegroups.com <mailto:raxml+un...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/raxml/f8665e97-130e-428f-b764-
> df7d1fb666b9n%40googlegroups.com <https://groups.google.com/d/msgid/raxml/f8665e97-130e-428f-b764-
> df7d1fb666b9n%40googlegroups.com?utm_medium=email&utm_source=footer>.