The next Omarchy installer unpacks a prebuilt system image instead of installing about 940 packages one by one
The next Omarchy ISO will ship the shared part of every install as a ready-made btrfs image, so the installer only has to add the few packages that differ between machines. A follow-up change also makes new encrypted installs cheaper on the CPU.
Image: Omarchy (omarchy.org)
Installing Omarchy is about to get a lot quicker. On October 25, 2026, @emirb merged “Speed up installer (F1-style work)” into the quattro branch of omarchy-iso. It’s a 398-commit landing branch with a simple goal, in the PR’s own words: “making the installer feel closer to F1 pit stop.” The change is coming in the next version of the ISO.
The core idea, according to the PR: “the installer unpacks a prebuilt root image and adds this machine’s packages on top, instead of pacstrapping ~940 packages.” Until now, every install pulled the same large package set through pacman one package at a time. Now that shared set is built once, when the ISO is made, and simply unpacked onto your disk.
The design comes from Jonny Heggheim (@hegjon), and emirb’s PR gives him all the credit for it. In his original PR #113, hegjon explained that the roughly 940 packages are the same on every machine. Only the kernel, CPU microcode, audio firmware and Tailscale change from one host to the next. He found that pacman extracted that set single-threaded at about 40 packages per second, and that adding more cores didn’t help.
His fix builds the shared set into a btrfs subvolume at ISO build time and ships it as a compressed btrfs send stream. The installer receives that stream onto the target disk, swaps it in as the root subvolume, then runs a small per-machine pacstrap for whatever the image doesn’t include. Because btrfs stores the already-compressed data as it arrives, the PR says unpacking takes about the same time no matter how many cores you have.
The numbers in PR #113 come from hegjon’s branch in August, so treat them as a preview. On a 16-vCPU QEMU VM, the install dropped from 43 seconds to 24 seconds with the exact same 942 packages, and the installed system shrank from 5.73 GiB to 3.98 GiB. On a Ryzen 9 9950X workstation, a full interactive install went from 53 to 36 seconds. There’s a trade-off: in that test the ISO grew from 6.2 GB to 6.84 GB.
Encrypted installs get an extra tweak. PR #336, merged on October 26, 2026, formats new encrypted roots with 4096-byte encryption sectors instead of 512-byte ones. Before, every 4 KiB btrfs block went through eight cipher requests. Existing installs keep their current layout, since the PR notes that the sector size can’t change after formatting without rewriting the whole volume.
The gains depend on your hardware, and the PR says so. On emirb’s i5-12600K with a Samsung 990 PRO, large sequential reads went from 4,106 MB/s to 6,802 MB/s, but the install itself wasn’t any faster because that CPU has cores to spare. On a smaller machine it helps: hegjon measured a two-core ThinkPad T460s, where the installer’s image write dropped from 2.44 s to 2.09 s. On every machine, the PR adds, large cold reads get faster and each byte costs less CPU.
