data centers

Linux/Bash playbook to cut AI server power without hurting throughput: baseline with powertop and nvidia-smi; tune CPU governors/EPP and turbo; cap GPU power and lock clocks; pin jobs to GPU-local NUMA; persist settings via tuned and powertop; optionally use Intel RAPL; mind thermals and NIC/storage power states; automate profiles for stable, greener, lower-cost training and inference across your fleet.