Linux 7.4-rc1 posts record 19,671 commits as AI fixes join regular development

A record-setting merge window

Linus Torvalds released Linux 7.4-rc1 on November 1 local time, closing out the largest merge window in kernel history. According to a report on dev.to, originally published by Joho 3 Todai, the roughly two-week feature integration period collected 19,671 commits, well above the 15,000 or so that a typical window gathers.

The acceleration is visible across the last four cycles: 13,963 commits for 7.1, 14,395 for 7.2, 16,267 for 7.3, and now 19,671 for 7.4.

The raw diff is noisier than it looks

Torvalds described the window as big, but cautioned against equating commit volume with new features. About a third of the diff consists of AMD GPU register offset and bitmask definitions, the bulk header-file work required whenever a new GPU is brought up, which he suggested discounting as noise. With that set aside, the report characterizes the release as typical: roughly half the changes touch drivers, with the remainder spread across architecture updates, filesystems, tooling, and core kernel areas including networking, Rust, BPF and memory management. The source tree now stands at around 41.61 million lines.

AI fixes arrive as regular development

A second force behind the record is AI. In his release announcement, Torvalds said he prefers seeing the flood of AI-prompted work land during the merge window rather than arriving mainly as additional fixes during the release candidate period, a pattern he described as a "much healthier model". Kernel development splits into a two-week merge window for features followed by several weeks of release candidates for stabilization; if fixes for AI-discovered bugs cluster in the RC period, the stabilization phase itself becomes unstable.

Torvalds also noted that patches proposed by AI itself are increasing alongside fixes for problems AI uncovered, though he added that these patches are often rewritten before they are accepted.

Review load and the LTS incentive

The shift carries costs. According to the dev.to report, maintainers in the networking subsystem continue to be overwhelmed by the volume of AI-generated bug reports and patches, and the mismatch between the low cost of producing patches and the high cost of reviewing them was already visible during the 7.3 cycle. It is reasonable to infer that the pressure grew again in 7.4 as the commit count climbed to 19,671.

Torvalds himself cautioned that AI is not the only reason for the size. He raised the possibility that developers, anticipating 7.4 would become this year's long-term support kernel, brought changes forward so they would land in a release with extended support.

New hardware support from AMD to Apple

The release also lays groundwork for a broad set of next-generation hardware. On the AMD side, preparations for Ryzen Zen 6 are in place, a PerfOpt feature aims to make AI and LLM workloads more efficient on Radeon integrated GPUs, and HDMI 2.1 output is now enabled by default in the AMDGPU driver. Intel gained support for the IPU8 image processing unit in its next-generation Nova Lake laptop SoC, and support for laptops powered by Qualcomm's Snapdragon X2 has expanded.

Initial Apple M4 support code has landed, though not to the point of everyday use, and mainline support has been added for Google's Tensor G5 SoC and Pixel 10 devices. Elsewhere, EXT4 gained buffered IOmap for routing buffered I/O through the iomap infrastructure, kernel build times were reduced, and the BFS filesystem, present in the tree since 1999, was removed.

What happens next

With rc1 out, the cycle moves into bug fixing and stabilization through successive release candidates. Torvalds said he is still considering two pull requests that might be merged in rc2. If development goes smoothly, 7.4 could see its final release in December and become this year's LTS kernel, though neither the date nor the designation has been confirmed.

Why it matters

The kernel is among the most heavily reviewed open-source projects now absorbing AI-generated code and bug reports at scale, and 7.4-rc1 is a checkpoint in that transition. Torvalds's framing, with AI work arriving as regular merge-window traffic rather than a late flood of fixes, signals that the community's preferred path is normalization rather than resistance. Whether the assessment holds will become clear during the RC period: a manageable fix rate would validate the model, while a spike would suggest AI output is simply being deferred. The binding constraint is reviewer capacity, not code production. For users, the release also primes mainline Linux for the next wave of laptop and desktop silicon, and a December LTS landing would turn that groundwork into a stable base for years.