Bug in LinearAlgebra + Plots-PyPlot combo on intel i9-9900?

Are you running the 32-bit Julia?

I’m struggling to understand how there could be a conflict(?) between Julia’s OpenBLAS and NumPy’s BLAS(?) on a 64-bit machine, since on 64-bit machines all of Julia’s OpenBLAS functions have a 64_ suffix (not used by NumPy’s BLAS) precisely to avoid conflicts.

1 Like

I’m running latest Windows 10 64bit, with 64bit Julia. Here is the downloaded file from julialang.org:
image

One of my students who has better IT background than me, suggested that I add a System Environment variable:
image

I don’t know if that helped, though.

I am using the miniconda installed by Julia and I have no other python distributions installed.

Miniconda installed by julia should be using MKL on my machine.

@BLI Can you test this issue on your machine?

https://github.com/JuliaLang/julia/issues/36976

If you can reproduce the above issue, then maybe PyPlot is not related.

@BLI. It turns out that issue in the above post can be reproduced by someone else, so it is not related to PyPlot+LinearAlgebra here.

Hm…

julia> x=zeros(ComplexF64,10001);

julia> x'*x
0.0 + 0.0im

This is while run on Julia 1.4.2, but on one of my i9-9900 machines which gives the buggy rank response.