try running benchmarks on self hosted

[SimonDanisch]

No description provided.

[MakieBot]

Benchmark Results

SHA: 7157d8c21b170eaacdaef2c3feefe639cd017dd4

Warning

These results are subject to substantial noise because GitHub's CI runs on shared machines that are not ideally suited for benchmarking.

[SimonDanisch]

Two take aways:

1. It's suprising how similar the noise is for a complete dedicated machine, that has nothing else to do but run the benchmark on linux, even with the tips from BenchmarkTools applied : https://juliaci.github.io/BenchmarkTools.jl/stable/linuxtips/.
   We probably still need to make sure that we avoid thermal throttling (it's a mobile chip).
2. It's crazy how much faster a dedicated machine is (from 25 min to 5min CI runtime), likely because the whole julia folder stays, without a lengthy up/download for the action cache and I guess a few other factors, like better CPU speed and GPU support.

I will close this for now, and come back to it later, when we have a dedicated machine for CI (still needs some containerisation + a stable location).