FYI here’s a follow-up comment on the FoldsCUDA.jl link: Advice for improving Monte-Carlo code - #25 by tkf