Re-posted from: https://bkamins.github.io/julialang/2023/12/01/plot.html
Introduction
Makie is a plotting ecosystem for the Julia language that is extremely feature-packed and actively developed.
Recently its core package Makie.jl reached version 0.20. Its core developer Simon told me that
the package now loads much faster than it was the case in the past.
The “time to first plot” issue is often raised by new users of the Julia ecosystem as important.
Therefore a lot of work was put both by Julia core developers and by package maintainers to reduce it.
In this post I thought that it would be interesting to check how CairoMakie.jl compares to Plots.jl.
The Plots.jl package is another great plotting ecosystem for Julia. It is more lightweight, so in the past it was seen
as faster, but less feature rich. Let us see how the situation stands currently.
From the Makie ecosystem I have chosen CairoMakie.jl as I typically need 2-D production-quality plots.
The code in this post was tested under Julia Version 1.10.0-rc1, CairoMakie.jl 0.11.2 and Plots.jl 1.39.0.
Installing the package
This, and the following tests, are done in separate Julia sessions in separate project environments
for both plotting ecosystems.
The first timing we make is package installation. The results are the following:
- CairoMakie.jl 0.11.2: 241 dependencies successfully precompiled in 227 seconds
- Plots.jl v1.39.0: 62 dependencies successfully precompiled in 78 seconds
CairoMakie.jl takes three times more time to install and has four times more dependencies.
Installation is one-time cost, however, there are two considerations to keep in mind:
- New users are first faced with installation time so it is noticeable (this is especially relevant with Pluto.jl).
- Since CairoMakie.jl has many more dependencies it is more likely that it will require recompilation when any of them gets updated.
Loading the package
After installing packages we can check how long it takes to load them:
julia> @time using CairoMakie
3.580779 seconds (2.75 M allocations: 181.590 MiB, 4.39% gc time, 1.71% compilation time: 49% of which was recompilation)
vs
julia> @time using Plots
1.296026 seconds (1.05 M allocations: 73.541 MiB, 6.00% gc time, 2.19% compilation time)
CairoMakie.jl takes around three times more time to load. This difference is noticeable, but I think it is not a show-stopper in most cases.
Having to wait 3.5 seconds for a package to load should be acceptable unless someone expects to run a really short-lived Julia session.
Simple plotting
Now comes the time to compare plotting time. Start with CairoMakie.jl:
julia> x = range(0, 10, length=100);
julia> y = sin.(x);
julia> @time lines(x, y)
0.559800 seconds (476.77 k allocations: 32.349 MiB, 3.13% gc time, 94.66% compilation time)
julia> @time lines(x, y)
0.012473 seconds (32.40 k allocations: 2.128 MiB)
vs Plots.jl:
julia> x = range(0, 10, length=100);
julia> y = sin.(x);
julia> @time plot(x, y)
0.082866 seconds (9.16 k allocations: 648.188 KiB, 97.32% compilation time)
julia> @time plot(x, y)
0.000508 seconds (484 allocations: 45.992 KiB)
The situation repeats. CairoMakie.jl is visibly slower, but having to wait 0.5 second for a first plot to be sent to the plotting engine is I think acceptable.
Note that the consecutive plots are much faster as they do not require compilation.
Conclusions
Given the timings I have gotten my judgment is as follows:
- CairoMakie.jl is still visibly slower than Plots.jl.
- Yet, CairoMakie.jl is in my opinion currently fast enough not to annoy users by requiring them to wait excessively long for a plot.
I think Makie maintainers, in combination with core Julia developers,
have done a fantastic job with improving time-to-first-plot in this ecosystem.
I can say that I decided to switch to Makie as my default plotting tool for larger projects.
However, I will probably for now still use Plots.jl in scenarios when I just want to start Julia and do a single quick plot
(especially on a machine where it has to be yet installed).