FAT32 is thirty years old and is still in almost everyone's pocket. It is on the USB drives passed around between offices, in camera cards, in the boot partitions of UEFI machines and in the file system a handheld console expects to find in order to read a games library. Nobody chose it for performance, and that is precisely why the improvement the 7.5 kernel is preparing is so striking.
The patches already under review boost their performance by up to 4.06 times in testsThe figure is not an average or a marketing promise; it is the best case measured from a round of changes that tackles the system's historic bottleneck, which is the way it locates free space to write. It also arrives without a format change, so a card formatted a decade ago benefits as well.
In the same development window, the kernel received patches to support AirDrop with Macs carrying the T2 chip and began preparing the removal of the last code optimized for MMX processors. These are three different pieces of work that share one characteristic: none of them will make a launch headline.
A new lease of life for the old FAT32
The problem with FAT32 has always been space allocation. The table describing which clusters are free does not have a structure designed for fast lookups, so writing a large file meant traversing it frequently, and the cost grew with the size and fragmentation of the volume.
The changes in 7.5 reorganize that search to take advantage of disk locality and reduce full traversals. The result depends on the write pattern, the card's size and how fragmented it is, which is why the improvement is announced as a maximum of 4.06 times rather than a fixed value.
The improvement arrives without changes to the format, something that matters more than it seems. A FAT32 volume created in 2008 benefits just as much as one formatted yesterday, and there is no migration or loss of compatibility with cameras, televisions or audio equipment.
4 GB limit
The scenarios with the most headroom are those involving many small files or large volumes with little free space. Copying a photo session to an SD card, dumping a video project to an external drive or writing a system image to a USB stick are the tasks where the difference is noticeable. On desktops the improvement will be invisible most of the time, because the user rarely writes more than a few hundred megabytes in a row, and those preparing cards for cameras will notice that the progress bar stops lying.
The other side is that the improvement does not touch reading. FAT32 remains a system without a journal or permissions, with a 4 GB limit per file and with names that behave differently depending on the operating system reading them, so the announcement should not be read as an invitation to use it for backups.
AirDrop and the removal of the old MMX
The patches for AirDrop are another story. What was sent to the kernel development list allows a Linux machine with a T2 chip to find and receive files from Apple devices, a feature that until now required third-party tools and a fair amount of patience.
The state is explicit in the mailing list itself: these are patches under review, not code that will definitely be in 7.5, and they depend on Apple not changing the protocol. The feature does not cover all models either, because the T2 is only in one specific generation of machines and the rest of the Macs need other pieces.
For the Linux user on a Mac, the news is good and still not usable. For everyone else, it confirms a trend the kernel has been showing for two years now: Apple hardware support comes in through individual contributions and moves more slowly than the rest of the tree.
The third move is a cleanup. The kernel has been carrying MMX optimizations since the nineties, the instruction set Intel introduced with the Pentium MMX in 1997, and the patches preparing 7.5 remove the last ones left. Nobody with a processor from this century is affected, and the benefit is maintenance: every code path kept for compatibility with extinct hardware has to be tested and documented in every release.
How to test it without waiting for 7.5?
Anyone who wants to measure the improvement can do so with the current 7.4, because some of the changes are already in the stable tree, and compare with the previous version on the same machine and the same card. The test has to use the same write pattern, or the result will say nothing.
For a useful measurement it is worth writing a set of files of varying sizes instead of one giant file, and repeating the operation with the volume almost full. That second case is the one that truly reflects real-world use of a camera card.
Kernel 7.5 will arrive in 2027 according to the usual schedule, and there is no need to wait to decide whether to migrate a boot library to another file system. The answer is still the same as five years ago: if the device needs to read it from a console or a camera, FAT32 is the option.
The rest of the cycle comes with more maintenance work than headlines. 7.4 debuted a new tree dedicated to memory management and a second round of changes fixing old bugs that had gone unresolved for years, a pattern that already appeared when the kernel started booting 50 ms earlier with the new memblock.
The kernel's boring work
It is worth remembering why these improvements take so long to arrive. Kernel 7.4 was the cycle with the largest merge window in history, with half a million more lines of code, and even so the changes you notice on the desktop are still those in file systems and memory management, not new drivers.
For anyone who uses Linux every day, the practical consequence is that the FAT32 improvement will show up in an update they will probably install without reading the notes, and that they will only notice when copying photos. That is the usual fate of good infrastructure work, and also the reason it is worth occasionally looking at what is cooking in the merge window before it arrives.
The contrast with what ext4 changed in the same cycle helps explain the split: modern systems get architecture improvements, and old ones get algorithm improvements because there is no room to redesign them. Both paths end in the same place, which is a disk that responds better than the user expects.