- Posted on
- Featured Image
A Bash-first Linux tuning guide to cut AI training runtime by 10–30% without changing model code: measure with htop/iostat/perf and GPU tools, then optimize CPU governor and NUMA affinity, THP/hugepages and allocators (jemalloc/tcmalloc), I/O schedulers and readahead, and NVIDIA settings. Includes apt/dnf/zypper installs, concrete commands, a reversible script, and a test-measure-iterate workflow.