CERN leaves RHEL and moves its 2,200 control computers to Debian 13

Late August 2026. CERN announces, at the MiniDebConf in Winterthur, Switzerland, that it is moving its industrial accelerator-control computers to Debian 13. 2,200 machines. That is how many systems are in scope, targeted for end of 2026. Why it matters: the organization that co-maintained Scientific Linux for a decade is turning its back on the Red Hat line for one precise reason — the CPU baseline its distribution imposes, not licensing and not price.

A RHEL institution changes course

CERN, the European Organization for Nuclear Research, is best known for the Large Hadron Collider (LHC). But to long-time Linux users it is also a RHEL institution: the lab co-maintained Scientific Linux, the RHEL derivative, with other research bodies, ran it for about ten years, then switched to CentOS in 2015. A decade later, it is leaving that lineage for part of its infrastructure.

The scope of the move has to be read precisely. CERN has clarified that the migration concerns its industrial accelerator-control computers — the machines that drive the magnets, radio-frequency cavities and instrumentation of the LHC — along with their embedded systems. The data centers and experimental computing, by contrast, stay on RHEL and AlmaLinux. This is not a blanket rejection of Red Hat; it is a targeted decision about the class of hardware that is hardest to refresh.

The real trigger: the -march=x86-64-v2 flag

CERN engineers had considered CentOS Stream as a more natural path. But, in their MiniDebConf talk, “the straw that broke the camel’s back” was the default -march=x86-64-v2 compiler flag, described as “forced obsolescence” of older hardware.

The mechanism is technical but clear. A distribution that raises its CPU baseline requires a minimum instruction set to run at all: SSE4.2, POPCNT and a few others. RHEL 10 crossed that line in 2025 by moving to x86-64-v2, effectively dropping processors from before 2009-2011 — a generation that, inside a particle accelerator, still powers thousands of instrumentation boards designed to last twenty years. An accelerator does not refresh its control electronics the way a fleet of workstations gets refreshed: a board installed in 2007 may still have fifteen years of service ahead of it.

For CERN, choosing Debian 13 is therefore not ideological. Debian is the distribution that keeps a low, stable CPU baseline, which lets the same modern operating system run on old hardware without replacing it. The opposite decision — following the Red Hat line — would have forced a wave of hardware replacement with no functional payoff.

Why CentOS Stream was not the answer

CERN’s choice of Debian did not happen in a vacuum; it follows a decade of upheaval in the Red Hat ecosystem. Scientific Linux, the RHEL rebuild that CERN co-maintained, was discontinued in 2019. CentOS Linux, the free binary-compatible rebuild most labs migrated to, was killed in 2021 and replaced by CentOS Stream, a rolling preview that sits just ahead of RHEL. For an institution that needs a frozen, reproducible platform on hardware that must run unchanged for years, a rolling distribution is a poor fit.

The x86-64-v2 baseline is the second squeeze. RHEL 10, released in 2025, requires CPUs that support SSE4.2 and POPCNT — instructions that first appeared around Intel’s Nehalem generation in 2008 and AMD’s Bulldozer in 2011. In a workstation fleet that is refreshed every few years, that requirement is invisible. In an accelerator where a control board commissioned in 2007 can legitimately stay in service past 2030, it is a forced hardware migration. Debian 13 still boots on x86-64-v1 machines, which is precisely why it fits CERN’s control systems without replacing a single board.

What Debian gives, and what it costs

The move to Debian is not friction-free, and CERN does not pretend otherwise. Its engineers flag two main difficulties: a lack of standard tooling for automated package building and publishing, with gaps throughout the official toolchain, and tools that do not support multiple versions of the same package. For an organization that maintains its own packages and repositories, those gaps are a real engineering cost.

That is the flip side of any distribution migration: you trade one constraint — here, the hardware obsolescence imposed by the baseline — for another — packaging tooling that is less turnkey. CERN’s calculus is that the first constraint is structural and non-negotiable, while the second is solvable with engineering work. The deadline is tight, in any case: 2,200 machines to move to Debian 13 by end of 2026.

The lesson for long-lived fleets

This move reaches far beyond CERN. It illustrates a tension running through all of enterprise Linux: commercial vendors align their baselines with recent hardware for performance and maintainability reasons, while industrial, embedded and homelab fleets live on timescales where the hardware far outlives a distribution’s support cycle. The tipping point is the CPU baseline, not the price.

# Check whether a CPU supports the x86-64-v2 level (SSE4.2 + POPCNT required)
grep -o -E "sse4_2|popcnt" /proc/cpuinfo | sort -u

If the command returns both flags, the machine is x86-64-v2 compatible and can follow a RHEL 10 track. If either is missing, the machine is destined for a low-baseline distribution — Debian, or a derivative — or for replacement. Running this check across an inventory is the hardware equivalent of a software end-of-life audit: it reveals, years ahead of the deadline, which machines a future distribution upgrade will condemn.

It is worth noting where the baseline is heading next. x86-64-v3 and x86-64-v4 targets have been discussed for years, and the pressure to adopt them will only grow as vendors chase AVX2 and AVX-512 performance. Every future baseline bump is another round of forced refreshes for anyone holding old hardware. CERN’s decision to anchor on Debian now is, in effect, a bet that a conservative baseline will remain available — and it is a bet other long-lived fleets can copy.

CERN’s lesson is all the more striking because it comes from an institution that is anything but a hobbyist shop. When an organization capable of maintaining its own RHEL derivative concludes that a CPU baseline is reason enough to switch distributions, that is a signal any team running long-lived hardware should take seriously.

Verdict

If you run industrial, embedded or aging hardware, audit the CPU baseline of your current distribution and of its next major release before it forces a refresh on you. The sse4_2 / popcnt check costs one second per machine and reveals the scale of the problem years in advance.

If you are weighing RHEL against Debian for a long-lived fleet, understand that the decisive question is neither price nor philosophy, but the rate at which the baseline climbs: a distribution that raises its baseline fast condemns your hardware; a conservative one leaves the choice with you.

If you are already on CentOS or a derivative at the end of its road, CERN’s precedent shows that a migration scoped by machine class — industrial control here, RHEL compute there — is healthier than a uniform decision. The right unit of decision is the hardware’s lifespan, not the distribution’s brand.

References

• Phoronix — CERN Transitioning Industrial Computers To Debian After Being A Longtime RHEL Institution, September 2, 2026
• Linuxiac — Debian 13 Is Taking Over 2,200+ Control Systems Across CERN, September 2, 2026
• MiniDebConf Winterthur 2026 — Controlling CERN’s accelerators with Debian (video presentation)