[Software] Linux 7.3 Kernel: The Features That Matter for Gamers and Home Labs

Quick answer: Linux 7.3 was released on October 18,
2026. The headline changes for everyday users are better VRAM handling
on GPUs with limited video memory (good news for low-end gaming), faster
Btrfs and ext4 in specific workloads, a new sandboxing building block
(failfs plus the fchroot(2) system call), SMB
server support for Apple Time Machine, and several virtualization and
FUSE improvements for home labs. You don’t need to compile it yourself.
Ubuntu 26.10, released October 15, 2026, already ships a 7.3-series
kernel, and most other distros will pick it up through their normal
update channels.

We checked kernel.org and the Linux Kernel Newbies 7.3 changelog on
October 28, 2026. Where a claim comes from a
distribution’s own release notes, we say so. We have not benchmarked
7.3, so every performance claim here is the kernel developers’ own, not
ours.

Linux 7.3 at a glance

Item Detail
Mainline release Linux 7.3, October 18, 2026 (kernel.org)
Previous stable series 7.2, latest 7.2.13 (October 26, 2026)
Long-term kernels listed 6.18, 6.12, 6.6, 6.1, 5.15, 5.10
First major distro with a 7.3-series kernel Ubuntu 26.10 (October 15, 2026)

What matters for gamers

Better VRAM management
on low-end GPUs

This is the one gamers should care about. When your GPU doesn’t have
much video memory, apps compete for it. The old behavior could leave the
app that arrived last, even one with higher priority, unable to get
enough VRAM. It would end up using GTT, which is system RAM the GPU can
reach, and that’s slower.

Linux 7.3 lets the dmem cgroup controller handle this
better. Instead of backing off right away, the system tries to evict
unprotected buffers to make room for the protected one. The kernel docs
describe the goal as letting applications actually use the memory
protection they were given, without “ping-ponging” between them. Kernel
Newbies links the AMDGPU work as “Fixing AMDGPU’s VRAM management for
low-end GPUs.”

What this means in practice: if you run a game next to a browser,
video call or streaming software on a card with limited VRAM, the game
may be less likely to get pushed into slower memory. It only helps if
your setup uses dmem cgroup protections, which is something
desktop environments and launchers need to adopt. Don’t expect a free
frame-rate boost on day one.

New hardware support worth
knowing

The changelog lists a long driver section. A few items gamers may
notice:

  • 2026 Steam Controller support in the
    hid-steam driver.
  • MSI Claw HID configuration driver, for the MSI Claw
    handheld’s controls.
  • AMD graphics: a DRM “color format” property, GFX
    6–8 modifier support, initial DCN 6.0.0 support and support for newer
    SMU, PSP and GC IP versions.
  • Intel: early groundwork for upcoming platforms,
    including Nova Lake device IDs in a few drivers.
  • Wi-Fi and Bluetooth: a new Wi-Fi HaLow driver, an
    NXP Wi-Fi driver for IW61x, and more MediaTek and Realtek device
    support.

Driver support in the kernel doesn’t guarantee a finished desktop
experience. You’ll still need matching firmware and user-space packages
(Mesa, firmware blobs) for new GPUs.

What matters for
home labs and self-hosters

Faster Btrfs and ext4
in specific cases

Per the kernel changelog:

Filesystem Change
Btrfs Direct I/O recovers performance lost in earlier releases
Btrfs On non-SSD disks, removing a 1-jiffy delay in logging workloads
(such as fsync) lowers latency and gives about 5x the throughput
Btrfs fsync with the no-holes feature (default since btrfs-progs 5.15) is
about 5x faster
ext4 Better direct I/O parallel write performance, faster
fallocate(), and a buffered-write regression fix

These are the developers’ numbers for specific workloads. Whether
your NAS or database sees a similar gain depends on your drives and
workload. Spinning disks with fsync-heavy workloads, such as some
databases or sync-heavy backup targets, are the likeliest to see
something.

SMB server support
for Apple Time Machine

The in-kernel SMB server (ksmbd) gains opt-in support for Apple Time
Machine extensions. If you want a Mac backup target without running
Samba, this is a step forward. The same changes add per-share SMB3
encryption, SMB3 request replay support and SMB Direct RDMA encryption
support.

It’s new and labeled opt-in, so try it on a test share before relying
on it for backups.

Virtualization:
working-set tracking for VM memory

userfaultfd(2) gains read-write protection tracking
(UFFDIO_REGISTER_MODE_RWP) to follow the working set of VM
guest memory. The docs describe it as similar to, but faster than,
mprotect(PROT_NONE) plus a signal handler. It’s designed
for VM memory managers that want to evict cold pages to tiered or remote
storage. This is mostly for people building hypervisor tooling, not for
someone running a handful of VMs, but it shows where VM memory
management is heading.

FUSE gets faster with
io_uring

FUSE servers can now pass the kernel a buffer pool, which cuts the
memory needed by FUSE-over-io-uring. A related series adds zero-copy, so
a FUSE server can access client pages or page cache folios directly.
Zero-copy needs CAP_SYS_ADMIN and buffer pools. If you run
a FUSE-based filesystem (a cloud-storage mount or a union filesystem,
for example), the benefit depends on whether the project adopts the new
interfaces.

RWF_DONTCACHE for
block devices

Databases that open a raw block device can now use the
RWF_DONTCACHE flag per request. They keep data in the page
cache when that helps and opt out when it doesn’t, without the downsides
of direct I/O.

Better cgroup scheduling
and sched_ext

  • sched_ext sub-schedulers are complete. A cgroup can
    now have its processes managed by its own BPF-based task scheduler, and
    different cgroups can use different ones. This is aimed at advanced
    tuning, such as giving a game or a latency-sensitive service its own
    scheduling policy.
  • cgroup scheduling got improvements. The kernel adds
    several weight-distribution modes (up, max,
    concur and tasks), with concur as
    the default, described as the most precise but most expensive. Cgroup
    scheduling now uses a single runqueue. Tasks with short time slices
    should see better scheduling latency.

If you run containers or many services on one box, cgroup scheduling
is one to watch. We can’t say how big the effect is on your machine
without testing.

Security and
sandboxing: failfs and fchroot(2)

Linux 7.3 adds a new pseudo-filesystem called failfs,
the counterpart of nullfs. Where nullfs means
“there is nothing here,” failfs means “nothing is supported
here.” Nothing can be mounted on it, and absolute path lookups fail with
EOPNOTSUPP.

A new system call, fchroot(2), is the
file-descriptor-based version of chroot(2). A process can
change its root to failfs, which leaves it able to look up
files only relative to directory file descriptors it already holds. That
gives programs a way to drop their filesystem access completely after
setup. The changelog says an unprivileged task can do this if it has
no_new_privs set.

This is a building block for developers of sandboxes and container
runtimes. You won’t use it directly, but it may show up in the tools you
do use.

Other changes worth a
mention

  • Kernel threads are separated from PID 1. PID 1
    (systemd, for example) now gets its own separate filesystem state, and
    kernel threads are isolated in a nullfs mount, so
    pivot_root() can’t rewrite kernel-thread state.
  • binfmt_misc can use BPF. Both matching
    and interpreter selection can be delegated to a BPF program. Nix and
    similar distros can use it for relocatable binaries without wrapper
    scripts.
  • NTFS can read Windows System Compression (WOF)
    files on an optional basis.
  • Old filesystems removed. The pull request list
    includes the removal of the efs and freevxfs
    filesystems.
  • Rust keeps growing. Additions include netlink
    support used by Rust Binder and more abstractions.
  • Bluetooth adds support for the Shorter Connection
    Interval feature.

Should you upgrade?

Upgrade (or choose a 7.3 distro) if:

  • You have newer hardware that needs recent drivers, especially recent
    AMD graphics, new Wi-Fi chips or the new Steam Controller.
  • You run a Btrfs or ext4 system with fsync-heavy workloads.
  • You’re curious about ksmbd for Time Machine, FUSE
    zero-copy or sched_ext.

Stay put if:

  • Your system is stable and your hardware is already well
    supported.
  • You run a production server on an LTS kernel. Kernel.org still lists
    several long-term series, and they’re getting updates.
  • You depend on out-of-tree modules (some NVIDIA, ZFS or
    virtualization modules) that may not build against 7.3 yet. Check your
    vendor first.

How to get Linux 7.3

  • Ubuntu 26.10 (released October 15, 2026): Ships a
    7.3-series kernel. Per Ubuntu’s release notes, the kernel in the release
    images (7.3.0-9.9) is based on 7.3-rc6, because the final upstream
    release came after launch. It’s not the final 7.3 code, and Ubuntu’s
    updates will move it forward. Ubuntu 26.10 gets 9 months of maintenance
    updates.
  • Other distros: Rolling distros typically pick up
    new mainline kernels within weeks. When we checked, Arch’s
    linux package JSON still showed 7.2.11, last updated
    October 17, so 7.3 hadn’t arrived there yet. We didn’t verify Fedora,
    Linux Mint or other distros’ timelines, so check your distro’s package
    page.
  • Manual install: Mainline kernel builds exist for
    Ubuntu and Debian-based systems, but they’re unsupported by the distro.
    We’d only do this on a test machine.

Before any kernel upgrade, keep your old kernel in the boot menu so
you can roll back.

FAQ

When was Linux 7.3 released?

Kernel.org lists Linux 7.3 as the mainline release dated October 18,
2026.

Does Linux 7.3
improve gaming performance?

It improves VRAM management for GPUs with limited video memory by
using dmem cgroup protection and eviction. The developers
framed it as a fix for competing apps pushing each other into slower
system memory. We haven’t benchmarked it, and the benefit depends on
your desktop and launcher using the new protections.

Is Linux 7.3 a
long-term support kernel?

Kernel.org’s list didn’t show 7.3 as long-term when we checked. The
long-term series listed on October 28, 2026 were 6.18, 6.12, 6.6, 6.1,
5.15 and 5.10. Check the releases page for updates.

Which distro ships Linux 7.3?

Ubuntu 26.10 ships a 7.3-series kernel (based on 7.3-rc6 at release).
Others will follow on their own schedules.

Does Linux 7.3
support the new Steam Controller?

The changelog lists “steam: Add 2026 Steam Controller support” in the
input drivers. Full functionality also depends on user-space software
such as Steam.

Can I use ksmbd
for Time Machine backups now?

Linux 7.3 adds opt-in Time Machine support to the in-kernel SMB
server. It’s new, so test it before trusting your only backups to
it.

Will my NVIDIA or ZFS
setup work with 7.3?

We didn’t verify this. Out-of-tree modules often need updates for a
new kernel. Check your vendor or distro before upgrading.

Sources

All checked October 28, 2026.

By Early Adopted Editorial Team. Last updated: October 28,
2026

1 thought on “[Software] Linux 7.3 Kernel: The Features That Matter for Gamers and Home Labs”

  1. Pingback: [Software] Ubuntu 26.10 vs 26.04 LTS: Which One Should You Install? – Early Adopted

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top