I’ve looked for some solution, and I found the repository of JuliaComputing/MKL.jl and this topic " Unusually bad performance of eigen() compared to eig() in Matlab". After following the tutorial the code is running 3x faster than the first version, here is the benchmark
BenchmarkTools.Trial:
memory estimate: 379.85 MiB
allocs estimate: 825
--------------
minimum time: 9.068 s (1.35% GC)
median time: 9.068 s (1.35% GC)
mean time: 9.068 s (1.35% GC)
maximum time: 9.068 s (1.35% GC)
--------------
samples: 1
evals/sample: 1
Many thanks to @kristoffer.carlsson by the repository.