Julia's applicable context is getting narrower over time?

[tim.holy]

@lungben

Only memory allocations are much faster in Numpy than in Julia (10 µs vs. 1 ms for 1M Float64).

julia> using BenchmarkTools

julia> @btime Vector{Float64}(undef, 10^6);
  8.377 μs (2 allocations: 7.63 MiB)

Are you by any chance comparing initilaized vs uninitialized memory?

julia> @btime zeros(10^6);
  452.453 μs (2 allocations: 7.63 MiB)