Options for large-scale dense linear algebra

My team is trying to identify options for dense-matrix distributed linear algebra on ~100s-1000s of CPU cores spread over multiple nodes. We are hoping for at least QR and Cholesky factorization, as well as linear equation solvers.

Initially, we expected to use ScaLAPACK.jl, but it does not seem to be maintained actively. I see someone recommended Elemental.jl in this forum a few years ago - is that the best choice? Are there others?

1 Like

ScaLAPACK or PLAPACK? You’d need to interface with @ccall, however (and this is exactly what ScaLAPACK.jl tries (tried?) to do).