Performance

Here are some simple performance measures to get a feeling of the productivity gains, compared to my existing Lenovo machine (Intel 12 cores). Even though these metrics specific to my workflow, they are documented for anyone curious. The performance critical parts of my projects is around video editing, Rust compilation and development with Tauri.

Benchmark preparation

Install Rust toolchain (see rustup.rs) and Tauri prerequisites.

git clone https://github.com/samuelroland/dme
git clone https://codeberg.org/samuelroland/runox
cd runox/cli
cargo fetch

Run the benchmark

Full Rust compilation in release mode
```
cd runox
time cargo build -r
```
Full Rust compilation with Cranelift in debug mode
```
cd runox
time cargo +nightly build
```
Incremental Rust compilation in release mode. Rebuild after changing the text inside a println.
```
cd runox
sed -i "s/Configured saved/CONFIGURED SAVED/" src/main.rs
time cargo build -r
```

Rust recompilation in debug mode after a Tauri application change. We are using the fastest linker available Wild.

cd dme/app
pnpm tauri dev ../../README.md # this opens the README at the root of the repository
# Wait for the full compilation to finish
# Do a simple println text change and start a chronometer at the same time
# Wait until the desktop app is opened again and is showing the Markdown document
# Stop the chronometer

Video rendering in Kdenlive, with a 11 minutes film. Open an existing Kdenlive project and export the video to MP4.

Machines

The following machines were used to run this benchmark.

Fedora Intel, on Fedora 44 with Intel processor 12th Gen Intel(R) Core(TM) i7-1255U including 12 cores. No GPU.
On a Macbook Air M2 2022 including 8 cores and 8GB of RAM
- MacOS ARM: Running MacOS Sequoia 15.7.7
- Fedora ARM: Running Asahi Linux with Fedora 44 Asahi Remix

Results in july 2026

On Fedora Intel
- Full Rust compilation: 48.67s
- Full Rust compilation with Cranelift in debug mode: 16.58
- Incremental Rust compilation: 20.59s
- Tauri Rust debug recompilation: 15s
- Video rendering: 38m18s
On MacOS ARM
- RAM usage without activity: 3.55GB
- Full Rust compilation: 23.9s
- Full Rust compilation with Cranelift in debug mode: ??
- Incremental Rust compilation: 6.66s
- Tauri Rust debug recompilation: ??
- Video rendering: 4m58s
On Fedora ARM
- RAM usage without activity: 3GB
- Full Rust compilation: 18.02s, 17.68s
- Full Rust compilation in debug mode: 10.61s
- Full Rust compilation with Cranelift in debug mode: 9s
- Incremental Rust compilation: 6.45s
- Tauri Rust debug recompilation: 10s
- Video rendering: 5min58s with the Kdenlive Flatpak version, 5min24s with the Kdenlive DNF version (at 2026-07-12)

Discussion

Wow, the Rust improvement is very clear! For a full compilation going from 48s to an ARM machine with 23s on MacOS and finally 18s on Asahi Linux, this is awesome ! The incremental compilation is also very impressive, from 20s to 6.6s and finally 6.45s.

The Tauri recompilation feedback loop is reduced from ~15s to 10s. In practice, as my Lenovo machine is regularly freezing during this recompilation it was taking way more than 15s, until 1min22s sometimes... On Fedora ARM, no freeze was noticed for now.

Using the GPU on ARM is obviously faster than the Fedora Intel that is using only the CPU to do the rendering. It is interesting to notice that Fedora ARM is a bit slower than MacOS ARM (which takes one less). No further research was done to understand this difference.