When I use
which corresponds to (IMHO) the multicore peak perf of my PC.
In addition I see top reaching 800%.
parallel flag seems to be ignored.
Is it expected ?
You can see what
parallel does here: https://github.com/JuliaLang/julia/blob/5486cc223c2b92b45d7d5e8e9cdbca9ffa49ab07/stdlib/LinearAlgebra/src/LinearAlgebra.jl#L400-L414.
The matrix multiplication uses BLAS and will be multithreaded.
parallel determines if multiple julia workers should be started.