Steven Barrett shipped Liquorix Linux Kernel 7.2-5 today, the third incremental update in under thirty-six hours. The release focuses on Project-C scheduler optimizations that keep coupled tasks on SMT sibling threads and prioritize placing sync wakees on the waker's core before looking at idle CPUs. It's a niche but highly relevant update for Linux gamers and audio producers chasing lower wake latency, though the project remains AMD64-only and lacks official support for ARM hardware.

Another Liquorix Kernel Drops, And It’s Already a Third Version in Two Days

Project-C scheduler tweaks aim to keep related threads on SMT siblings, shaving precious microseconds off wake latency.

Steven Barrett shipped Liquorix Linux Kernel 7.2-5 today, sitting on top of upstream Linux 7.2.3. It's not just another stable merge. This one wraps two fresh Project-C scheduler patches aimed at how the kernel handles simultaneous multithreading, and it marks the third incremental scheduler refinement in roughly 36 hours.

Barrett published 7.2-3, 7.2-4, and now 7.2-5 between September 3 and 4, 2026. Each iteration chased the same goal: keep tightly coupled threads on SMT sibling cores instead of scattering them across the silicon.

What the new patches actually do

Both changes live inside sched/alt. The first tells the SMT balancer to recognize existing task pairs and stop tearing them apart during periodic load redistribution. The second rewrites the priority for synchronous wakeups, landing a freshly woken thread on the waker's sibling core before falling back to a random idle CPU. It's all about preserving L2 cache locality. SMT siblings share that cache. Keeping producer-consumer chains together cuts the cold-cache penalties that would otherwise spike latency.

What does this change for you?

If you run audio production software, games, or anything with tight thread coupling, the payoff shows up in measured microseconds. A buffer processor waking an output thread now sits right next to the waker. Game engines see the same treatment for physics-to-render chains. Desktop responsiveness improves too, since UI threads that spawn worker tasks won't lose heat in another core's cache.

The progression from 7.2-3 to 7.2-5 reveals a straightforward feedback loop. Barrett shipped an initial SMT-aware patch, tested it across workloads, reverted a sibling skip optimization that didn't translate cleanly, and then doubled down on sibling affinity. That kind of rapid iterate-test-refine cycle is rare outside major upstream kernel teams. It also explains why the changelog reads like a lab notebook.

The Liquorix ecosystem at a glance

The project has been around for over a decade, riding on Debian, Ubuntu, and Arch. It swaps the mainline CFS scheduler for the BMQ variant, which traces back to the old CK patchset and has been evolved by the community for 15+ years. Hard kernel preemption, 1000Hz tick rates, and a full interactive tuning profile ship by default. The build runs entirely on Docker-based CI and targets AMD64 only. You can drop it into your system with a single curl command or pull it from the official PPA and AUR.

It's not for everyone. The kernel trades throughput for interactivity, which means batch workloads and heavy compute jobs will see little benefit, and sometimes a step backward. You're also locking into a solo maintainer's pace. When Barrett ships, you get patches quickly. When he's busy, you wait.

You can grab the installer by running:

curl -s 'https://liquorix.net/install-liquorix.sh' | sudo bash

Head here for the full changelog.