After months of work I'm finally releasing Margine OS, my own atomic Linux distro, and the short version is that it's fast.
It's built on Bluefin DX, so Fedora bootc underneath, which means it keeps everything that already makes Bluefin nice to live with: it's atomic, every codec is in place, updates happen quietly in the background, and you can always roll back if something breaks. What I changed is mostly in service of speed. Instead of the stock Fedora kernel it runs the CachyOS kernel with the BORE scheduler, re-signed with my own key so it still boots cleanly under Secure Boot, and the installer walks you through enrolling that key so you never have to turn Secure Boot off.
Around that there are a few things I'd always wished for. You can switch the sched_ext CPU schedulers live from a small GUI (scx_lavd when I'm gaming, plain BORE the rest of the time). There's a little tool I wrote, Wayland Scroll Factor, for the touchpad scroll and pinch speed that GNOME stubbornly won't expose, which matters a lot since the Framework 13 touchpad is unusably fast without it. GNOME comes set up for tiling out of the box with o-tiling, a fork of System76's Pop Shell, plus Hyprland-style keybindings, and gaming is one command away with a native Steam/Proton stack, Bazzite-style. The whole image is built, tested and signed on CI, and the ISOs are distributed torrent-first through the Internet Archive.
I benchmarked the kernel honestly on the same laptop, a Framework 13 with a Ryzen 5 7640U, swapping only the ostree deployment between Margine OS and stock Bluefin DX: roughly 1.8x faster context-switch latency, +54% thread throughput, and 43 to 55% lower median scheduling latency, with a small cost at the worst-case tail, which is the expected BORE trade-off and honestly a sign the numbers aren't cherry-picked. The full method and raw data are on the site.
It's a personal, opinionated project with a single maintainer, so feedback and criticism are genuinely welcome. There's also an experimental NVIDIA variant I can't test myself, since I have no NVIDIA hardware, so if you run NVIDIA and feel like helping validate it, that would mean a lot.
Site and download: https://margine.the-empty.place/ Docs and the full benchmark: https://margine.the-empty.place/docs
Well, yes and no.
It’s true that the project originally started around my own hardware and personal needs. However, that mainly influenced choices such as the preinstalled Flatpaks, GNOME extensions, and default configuration. The work required to rebuild the image around the CachyOS kernel, automate the build process, sign the resulting packages, run smoke tests, and provide tools for switching CPU schedulers is not specific to my hardware. Other users can benefit from it as well. That is precisely why, after initially building it for myself, I worked on making the process reproducible and suitable for public distribution. The image can run on different Intel and AMD systems, and I have also created an NVIDIA image so that the project can be tested on hardware other than my own. Your point about the signing keys is fair: they are currently personal keys rather than keys managed by an established organization. This is still a small independent project, so it doesn't have the same governance or trust model as a large distribution. However, the entire build process is public, and users can inspect it or rebuild the image themselves.
As for whether it qualifies as a “distribution,” I agree that simply publishing an ISO as a torrent on the Internet Archive would not be enough. But that's not what defines the project. The project includes automated image and package builds, kernel integration, signing, testing, Secure Boot support, custom tools, and reproducible GitHub Actions workflows. Whether someone prefers to call it a distribution, a Universal Blue derivative, or a custom Fedora image is partly a matter of terminology, but it is certainly more than a manually modified ISO uploaded as a torrent. You can inspect the build history and the amount of automation involved here: https://github.com/daniel-g-carrasco/margine-image/actions
You’d have to pull out a lot of stuff to make a system that wouldn’t run on x86 64.
I think what you’ve done is cool and I was not trying to denigrate or minimize your work.
My intent was to help someone who seems like they don’t have 25 years of newsgroup discussion about what constitutes a distribution and why under their belt understand why it was perfectly reasonable and defensible to get all “woah there buddy” about it.