Status: Linux 7.3 is still in development. The stable release is expected in late October 2026. This article covers what has landed in the mainline tree so far and will be updated once the kernel actually ships. Sourced from Phoronix reporting, verified live on August 30, 2026.
- When Linux 7.3 actually arrives
- The scheduler fix that's the real gaming story
- The hybrid CPU fix for Intel Core Ultra laptops and desktops
- GPU driver changes: old cards, new cards, and vRAM overcommit
- The regression you need to know about if you run Nvidia
- Steam Controller and other input device support
- What's being dropped, and why
- Should you update now or wait
When Linux 7.3 actually arrives
The merge window, the roughly two-week period where new features get pulled into the mainline kernel, opened in mid-August 2026 after Linux 7.2's stable release. Based on the usual kernel release cadence, Linux 7.3 stable is expected in late October 2026. If you're on a rolling-release distro or comfortable running a mainline build, you can try release candidates before then; if you're on a distro with a fixed kernel version, expect it to arrive whenever your distribution packages the new stable series after release, which is often weeks later still.
The scheduler fix that's the real gaming story
The most significant gaming-relevant change this cycle isn't a graphics driver at all, it's a scheduler patch series nicknamed "flatten the pick," written by Intel engineer Peter Zijlstra. It reworks how the kernel's cgroup scheduling handles task priority, specifically fixing a case where heavy background load could badly hurt foreground application performance, exactly the scenario a game competing with background processes runs into.
The patch series was benchmarked on genuinely old hardware, an Intel Sandy Bridge CPU paired with an AMD Radeon RX 580, and showed a real FPS improvement along with meaningfully lower frametime variance under scheduling noise. That's not a synthetic benchmark chasing a big number; it's specifically testing the case this patch was written for, low-power or aging systems under real-world background load.
What that means practically: if you're running Linux gaming on older or budget hardware, or on a system where you keep background apps open while gaming, this is the kernel change most likely to produce a noticeable, measurable improvement for you specifically. If you're on high-end current-generation hardware with a clean, dedicated gaming session, the improvement will likely be smaller since you're less exposed to the scheduling contention this patch targets.
Worth understanding why this matters beyond the benchmark numbers: cgroup scheduling is how the kernel decides which process gets CPU time when several are competing for it, and the previous behavior could let background weight accumulate in ways that starved whatever you actually cared about, your game, of consistent scheduling slices. The "flatten the pick" naming refers to restructuring how that weight gets distributed across the scheduling hierarchy so a game's process doesn't lose out to background noise it shouldn't have to compete with in the first place. You don't need to understand the internals to benefit from it, but it explains why the fix helps most exactly when you have other things running, a browser with dozens of tabs, a background download, a Discord call, alongside your game.
The hybrid CPU fix for Intel Core Ultra laptops and desktops
Separately, Linux 7.3 includes cluster-aware scheduling improvements aimed at modern Intel Core Ultra CPUs with a mix of performance (P) and efficiency (E) cores. According to the reporting, this specifically fixes a load-balancing regression that had existed on hybrid CPU architectures for years, rather than introducing entirely new hybrid-CPU support. If you've been running a Core Ultra laptop or desktop on Linux and noticed inconsistent core utilization or scheduling oddities, this is the fix aimed at that exact problem.
GPU driver changes: old cards, new cards, and vRAM overcommit
The Direct Rendering Manager (DRM) subsystem pull for 7.3 covers a wide span of hardware. On the older end, GPU reset recovery has been improved for AMD's GFX7-generation cards, including the Kaveri and Hawaii chips, hardware that's a decade old at this point but still in real use on budget Linux gaming rigs. On the newer end, Intel's next-generation Nova Lake S graphics are being enabled by default, and next-generation AMD graphics IP blocks are gaining initial support.
The change most worth your attention if you run a card with limited vRAM is the DMEMCG eviction work led by Valve engineer Natalie Vock, improving how the AMDGPU driver handles video memory eviction when a game tries to allocate more vRAM than you physically have. In testing cited in the reporting, a game (Indiana Jones: The Great Circle) pushed to settings that tried to consume roughly 9GB of vRAM on an 8GB card remained playable even with 1GB effectively over-committed. That's a real, concrete improvement for anyone running an 8GB card against increasingly vRAM-hungry modern titles, and it's ongoing work, meaning further improvements are expected in later kernel versions beyond 7.3.
The regression you need to know about if you run Nvidia
Early Linux 7.3 benchmarking has also turned up a real problem worth flagging clearly rather than burying under the good news: notable regressions in graphics benchmarks specifically on the Nouveau driver, the open-source Nvidia driver. This was caught during early-cycle testing, which means there's time for it to be fixed before the stable release, but if you're running Nvidia hardware on the open-source driver rather than Nvidia's proprietary driver, this is a reason to hold off testing 7.3 release candidates for gaming until the regression is confirmed resolved.
Steam Controller and other input device support
The HID (Human Interface Device) subsystem pull brings the most directly gamer-facing changes in this release. Valve contributor Vicki Pfau's patches bring the 2026 Steam Controller up to feature parity with the level of driver support the original 2015 Steam Controller has had for years, and separately fix a long-standing gap in sensor support for the original controller itself.
| Device | What changed in Linux 7.3 |
|---|---|
| 2026 Steam Controller | Initial driver support, brought to parity with the original 2015 model |
| Original (2015) Steam Controller | Long-standing sensor support gap fixed |
| MSI Claw Gamepad | New dedicated HID driver added |
| Logitech devices | HID++ 2.0 reprogrammable button support |
| HyperX QuadCast 2 (USB gaming mic) | New driver for microphone mute status reporting |
| AMD Sensor Fusion Hub devices | Tablet-mode switch support added |
What's being dropped, and why
Every kernel cycle drops some old hardware support, and 7.3 continues a pattern worth knowing about if you're running genuinely old machines: old SGI drivers, some old serial card drivers, the FreeVxFS filesystem driver, and the EFS filesystem are all being removed. According to Phoronix's reporting, this driver removal trend has been accelerated in 2026 partly by the maintenance burden of sorting through a growing volume of AI/LLM-generated bug reports and patches, which makes maintainers less willing to keep carrying drivers with no active users. If you're running specialized old hardware that depends on any of these, it's worth checking before you update to a 7.3-based kernel down the line.
Should you update now or wait
If you're comfortable running release candidates and want to test the scheduler and vRAM improvements early, and you're not on Nvidia's open-source Nouveau driver, trying a 7.3-rc build once release candidates start circulating is reasonable. If you're running Nvidia hardware on Nouveau, or you just want a stable, boring gaming setup, waiting for your distribution's packaged stable release after the expected late-October debut is the safer call. Either way, treat everything above as merge-window reporting, not a finished feature list; kernel development can still change or drop things before the final release.
If you do want to try a release candidate
- Don't replace your working kernel. Install the release candidate alongside your current one rather than over it, so you can boot back into your known-good kernel from your bootloader menu if something breaks.
- Check your distro's mainline kernel tooling first. Ubuntu-based distros have the Mainline Kernel installer, Arch-based distros have AUR packages tracking upstream RCs, and most distros document a safe way to test an RC without touching your default boot entry. Use that instead of compiling from source unless you specifically want to.
- Confirm your GPU driver situation before you boot into it. If you're on Nvidia's proprietary driver, this Nouveau-specific regression doesn't apply to you. If you're on Nouveau, hold off until the regression is confirmed fixed.
- Test with something you'd notice a regression in. Run a game you know the normal frame time and FPS behavior of, not a new title where you have nothing to compare against.
None of this requires being a kernel developer. It's the same caution you'd apply to any beta software: keep a working fallback, and don't test something you can't afford to have break on your only machine.
