Apple's AI Engine Gets Smarter at Saving Memory
Your Mac could run AI faster and smoother without slowing down.
mlxrunner adapts saved history to memory pressure. It retains Metal prefix history using observed memory headroom and execution workspace rather than a fixed 8 GiB allowance.
When system compression grows, it releases unused buffers and cold history while preserving reusable paths and active requests. The changes cover pressure reclamation, snapshot ownership, memory observation, and cacheless request isolation.
- mlxrunner now uses available memory instead of a fixed 8 GB limit, reducing waste.
- It frees up unused memory when your Mac is under pressure, keeping apps responsive.
- This update is in testing (rc2) and will roll out to apps using Apple's AI engine.
Why It Matters
Less memory waste means smoother AI apps and fewer slowdowns on your Mac.