Linux 7.4-rc1 vs Linux 7.3: Scale, Speed and the Real Upgrade Call | ItInsightsWorld
Short answer first: a release candidate should not change what runs on your work machine this week. A few days ago Linux founder Linus Torvalds published Linux 7.4-rc1, the first test build of the Linux 7.4 series, which closes a two-week merge window and moves the Linux kernel into repair-and-harden mode. What makes this cycle worth reading about is not the "rc1" label but the volume behind it — Torvalds himself called it a "very big (BIG)" window — and the fact that several user-visible wins (AMD graphics, faster builds, quicker booting on giant servers, support for hardware that has not shipped yet) landed inside it. Below we line Linux 7.4-rc1 up against the stable 7.3 line you are probably running today, and say who should install now, who should wait about two months, and who can safely ignore the news.
60-second verdict before the details
- Everyday desktop user with a machine that works: stay put. The finished release is projected for late December 2026 — reporting from 9to5Linux and Linux Compatible points at around December 20, or December 27 if an extra candidate is needed. That version, not rc1, is your install target.
- AMD APU, handheld or new-laptop owner: this is the cycle with the gains you would actually notice, from AMD PerfOpt for integrated graphics to HDMI 2.1 output being enabled by default.
- Fleet, server or embedded planning: the real question is the long-term support (LTS) label. Multiple outlets report that Linux 7.4 is the likely next LTS, with Greg Kroah-Hartman taking over maintenance in early 2027. Nothing is officially stamped yet, but it is the version to plan around, not to fear.
Head to head: Linux 7.3.1, Linux 7.4-rc1 and the future Linux 7.4
| Dimension | Linux 7.3.1 (released recently) | Linux 7.4-rc1 (out a few days ago) | Linux 7.4 stable (projected) |
|---|---|---|---|
| Purpose | Maintenance repairs on the shipped line | First feature-complete test build; regressions expected | Polished result after a weekly repair cycle |
| Change volume | Small and targeted | About 20,000 code commits in one window, against a normal cycle of roughly 15,000 or so | Same feature delta, debugged |
| Risk profile | Lowest available | Highest of the three by design | Falls week by week toward release |
| Who picks it up first | Arch Linux-based distributions publish matching upgrade packages quickly | Fedora, known for chasing new kernels, and kernel developers | The next LTS wave, including Android and Chromebook class devices over time |
| Best fit | Machines that must keep serving | Spare box, test VM, new-silicon bring-up | Long-lived deployments and product roadmaps |
Why the headline number needs a second read
Torvalds described this merge window as "very big (BIG)" and named AI-related development as one important driver. His reasoning is a quality argument rather than a growth one: AI-driven work that lands during the window is more satisfying than the same work arriving later as an avalanche of urgent fix patches. In the test phase AI-related repairs will still show up, but the expected impact is smaller than in previous cycles.
The raw count also deserves unpacking. Torvalds reported about 20,000 code commits, where a typical window settles at roughly 15,000 or so; Phoronix's tally, cited by AI Weekly and HWBusters, counted 19,671 changes for this window against 16,267 in the Linux 7.3 window, about 21 percent higher. But a large slice is paperwork: Torvalds noted that about one-third of the patch volume is really AMD GPU hardware register definitions — offset tables and bit masks imported into the kernel, which inflate the diff log without changing behaviour on their own. Judge the release by the delta you can feel, not the delta you can count.
What you actually gain: the testable list
- AMD integrated graphics and local AI. The AMD PerfOpt change lets a privileged integrated GPU reach system memory with most IOMMU address translation skipped, keeping only basic read/write permission checks. Benchmarks reported by Phoronix measured 18 to 23 percent uplift for local large-language-model work on Strix Halo class APUs, while flagship parts with large carved-out memory saw far smaller typical gains of a few percent. The trade: a bit of DMA isolation is spent to buy latency, and users can force the behaviour off through an amdgpu module parameter.
- AMD display output. HDMI 2.1 support moves to enabled by default for Radeon hardware in this cycle, with variable refresh rate and auto low-latency mode added, closing a long-standing gap between Linux and Windows on modern TVs and monitors.
- Support for chips that are not on shelves yet. Preparation for the AMD Ryzen Zen 6 client platform lands ahead of the processors themselves, including BTB CTX isolation, RMPOPT optimisations and early P-State driver work, alongside Intel Rugged Panther Lake enablement. If you are bring-up testing next year's hardware, this release window is the baseline.
- Laptop and handheld specifics. Vendor driver updates in this cycle cover mainstream brands with hardware monitoring, fan control and battery charge limiting, plus handheld features such as fan-curve and battery reads on devices like the AYANEO 2S and OneXPlayer 3 — the kind of change that never shows in a feature trailer but decides whether a machine is pleasant to live with.
- Speed where it compounds. Coverage from Linux Compatible and Acta Verum describes a boot-time memory operation dropping from roughly 268 seconds to about 50 milliseconds on very large machines, with idle workqueue flushing reported 10 to 100 times faster. Phoronix has reported build-time work that could make full kernel builds around 36 percent faster and incremental builds about 70 percent faster. For anyone compiling kernels or booting high-core-count servers, this is the most immediately measurable line item.
Beyond the highlights, the first candidate of the Linux 7.4 series also carries routine breadth: drivers, processor architecture support, file systems, development toolchains and core kernel components — including storage and file-system tuning that reviewers have flagged in Btrfs and EXT4 work, plus early enablement for Apple silicon machines that is explicitly not yet end-user ready.
What you risk: the other side of a record window
Torvalds explained that a change in schedule caused by time off left two pull requests still unprocessed in his inbox. Rather than delay the first candidate, he shipped Linux 7.4-rc1 anyway, so whether that code joins Linux 7.4 at all may be settled at rc2. For anyone evaluating this like a product decision, that is the honest uncertainty clause: the feature set is nearly final but not quite.
Scale cuts both ways. More merged work means a wider surface for regressions, and the AI-driven influx adds review pressure — one anti-noise mechanism introduced for this cycle was reportedly reverted almost immediately, which tells you the tooling problem is being solved in public. There is also a supply-side motive: many developers may expect Linux 7.4 to become the next long-term support (LTS) kernel, so each of them wants their own work inside that version, which is exactly why future candidates may stay unusually large. Over the next few weeks the job is stabilisation and polish, and the risk curve falls with every candidate.
Decision matrix: match your situation to the right move
| Your situation | Recommended move | Why |
|---|---|---|
| Production server, NAS or shop terminal | Stay on your current kernel | rc1 exists to surface bugs, not to add uptime; wait for the stable release and, better, for the LTS designation to be confirmed |
| Desktop that runs fine on the 7.3 line | Upgrade when Linux 7.4 ships, not before | You get the same features with the rc-phase repairs already folded in |
| AMD APU laptop or gaming handheld on a rolling release | Install the candidates now and benchmark | PerfOpt and HDMI 2.1 defaults target precisely this hardware, and early adopters are the ones who report regressions |
| Building a product around next-generation x86 hardware | Track Linux 7.4-rc1 to stable | Enablement for upcoming platforms landed in this window, so early bring-up has no older base to use |
| Kernel developer, packager or CI owner | Rebuild your toolchain against rc1 today | Build-time and boot-time improvements pay back on every iteration |
| Rugged, aerospace or industrial design | Wait, then adopt as the long-lived base | Support for hardened SoCs is new here and will mature through the LTS period |
Four common misconceptions
- "rc1 means the project slowed down." It means the opposite: feature merging has closed and everything is now visible in one tree you can test.
- "20,000 commits equals 20,000 new features." It does not; about one-third of the patch volume in this release is AMD GPU register definition data.
- "LTS means it installs itself." LTS is a maintenance promise about longevity. Whether your distribution adopts that kernel is a separate product decision.
- "If I skip this cycle I skip a year of Linux." Maintenance updates continue — Linux 7.3.1 appeared recently — so you lose new hardware enablement, not security fixes.
Action checklist for the next about two months
- Confirm a rollback path — snapshot, spare kernel entry in the boot loader, or a reinstall stick — before touching any candidate.
- Install Linux 7.4-rc1 only on a spare machine or virtual machine, never on the device you depend on for work.
- Follow the cadence: convention gives roughly two months and seven to eight candidates, released every week, before the stable Linux 7.4 arrives.
- Watch rc2 closely for the two outstanding pull requests, which will tell you how final the feature list really is.
- On AMD integrated graphics, benchmark your own workload before and after; keep the module parameter that disables the new behaviour in your notes.
- File useful bug reports: kernel version, hardware model, and the relevant boot log beat a one-line complaint.
- For fleets, decide now whether the next long-term support (LTS) baseline will be this version and schedule migration for 2027 rather than reacting in December.
- If you want today's fixes without tomorrow's features, stay current on the 7.3 line; Arch Linux-based distributions push these updates fast and Fedora is usually among the first to deliver newer kernels.
FAQ: the questions readers actually ask
Is Linux 7.4-rc1 safe on my daily laptop?
It is a test build meant to be broken open by volunteers. Fine on a spare machine or VM; not the choice for the one device you cannot afford to lose an afternoon on.
When is the finished version due?
By Linux kernel convention, roughly two months of weekly candidates — usually seven to eight of them — precede the stable release, which current reporting places around December 20 or December 27, 2026.
Has the LTS label been confirmed?
No. Torvalds flagged that many developers appear to expect Linux 7.4 to become the next long-term support (LTS) release, and coverage by Greg Kroah-Hartman from early 2027 has been reported, but treat it as a well-informed expectation until the stable release ships.
Why does AI keep appearing in the story?
AI tooling now both generates patches and finds bugs at scale. Torvalds' position is that this work is healthier landing inside the two-week merge window than arriving late as emergency fixes; expect AI-related repairs in the candidate phase with less disruption than in the past.
Which distribution gets it to me first?
The fast-moving ones. Fedora tends to deliver new kernels unusually early, Arch Linux and its derivatives roll quickly, while Ubuntu-style releases pin to an earlier kernel — IT Home reported, for example, that Ubuntu 26.10 targeted the 7.3 series.
Bottom line
Put the three options on one scale: Linux 7.3.1 is the stability you already own, Linux 7.4-rc1 is the widest feature surface of the current cycle priced in regression risk, and the coming stable release is the version worth scheduling work around. Read the record-size numbers with the AMD register tables subtracted, test the performance claims on your own hardware rather than trusting the changelog, and let the long-term support question settle before you commit a fleet. That sequence turns a noisy kernel milestone into a calm, defensible upgrade decision.
Written by James Hartley for IT Insights. Questions or corrections: contact@itinsights.com.