Threads in Pluto

[r-acad]

I start Julia from the command line with julia -t 32 and I see 32 threads (Threads.nthreads() returns 32 in the REPL)

I then start Pluto and I only see 16 (Threads.nthreads() returns 16 when run in a Pluto cell)

Then I solve a linear system in the REPL (A\b, where A is a dense matrix and b a dense vector) and I see all the cores running in parallel in the resource monitor (Win10), but when I do this within Pluto I can only see one core working in this case.

In my code some sections seem to run in parallel and some others (which should be multithreaded) run on a single core.

Is this a limitation of Pluto or I am doing something wrong?

I am running Julia 1.6.1 and Pluto 0.14.5

Raul

[fonsp]

The Pluto server starts a new terminal for each notebook, and you specify the number of threads for notebook processes with:

Pluto.run(threads=16)

Using julia -t 32 to run Pluto does not work because it sets the number of threads for the Pluto server, not the processes that the server creates.

[r-acad]

Fons, you are my hero 🙏

[lungben]

Even without explicitly setting threads, Pluto uses a sensible default value (I think it is logical cores / 2 to account for hyperthreading - using more threads than physical cores usually gives a slow-down).