Linux 7.5 Set to Delete Last MMX Code, Ending 30-Year Pentium Era

A patch expected to merge into mainline Linux 7.5 will remove the last remaining Intel MMX-optimized software from the open-source kernel, retiring 1997-era Pentium instructions roughly three decades after their debut. The change strips MMX fast paths from the kernel's software RAID XOR code and cuts more than 800 lines of code down to just 12.

The kernel carrying the deletion is expected to be released in 2027, as Phoronix explains. By then, almost nobody will be running modern Linux builds on Pentium MMX or Pentium II hardware, so the practical cost of the removal is close to zero. Determined late-1990s Pentium devotees will still be able to use software RAID; it will simply lack the MMX fast path.

How much MMX was left in Linux?

Admittedly, there was not much left. Most MMX-optimized code disappeared from the operating system roughly half a decade ago, when developers removed MMX-optimized memory management code. This final patch is a classic case of Linux code housekeeping rather than a performance shift anyone will notice on current hardware. The takeaway is simple: the kernel sheds dead weight, and modern users lose nothing.

What was MMX, and why did it matter?

Intel introduced MMX in Pentium chips in January 1997 and marketed it as a MultiMedia eXtension for processor architectures in new PCs equipped with CD-ROM drives. The instructions arrived at the moment when consumer multimedia, CD-ROM software, and early video playback were driving PC purchases.

The name itself carries a curious legal history. Though the processors were marketed as enhanced by MultiMedia eXtensions, Wikipedia insists that "MMX is officially a meaningless initialism trademarked by Intel." In a subsequent trademark battle with longtime foe AMD, a document produced in court indicated that MMX stood for "Matrix Math Extensions." AMD leaned on this to sidestep the trademark, because, according to the encyclopedic source, an initialism cannot be trademarked. In plain terms, the name was marketing first and technology branding second.

Who actually shipped MMX?

After MMX's launch, all rival x86 processor makers integrated the instructions so as not to be left behind during the multimedia boom. A surprisingly large number of non-Intel chips ended up offering MMX support:

• Intel's Pentium MMX processors
• AMD
• Centaur
• Cyrix
• Rise
• Transmeta

That list reads today as a roster of the x86 market's vanished challengers, with only Intel and AMD still standing as volume x86 vendors. The takeaway: MMX became a universal x86 feature long before it became obsolete.

What came after MMX?

The instruction set's commercial success triggered a rapid succession of successors. AMD pioneered the 3DNow! instruction extensions, and roughly a year later Intel fired back with SSE and SSE2. Those streaming SIMD extensions superseded MMX as the standard path for multimedia and floating-point work, and the industry moved on. The MMX paths that remained in the Linux kernel were fossils of that transition rather than living optimization targets.

Why remove code nobody uses?

The mechanism here is straightforward maintenance economics. Every specialized code path in the kernel must be compiled, tested, and understood by maintainers, even if no realistic hardware exercises it. Deleting an 800-line MMX implementation in favor of a 12-line generic replacement reduces that maintenance surface permanently, with no measurable performance penalty on any machine likely to run kernel 7.5.

The change also marks a symbolic boundary. When Linux 7.5 ships, expected in 2027, the kernel will no longer contain any code written to exploit a Pentium-era feature, closing a chapter that began in January 1997. Expect further legacy-x86 pruning in future kernel cycles as maintainers continue clearing paths that only museum-grade hardware could ever reach.