gpu optimization

Explores how to maximize AI workload performance in containers: pick slim GPU‑ready base images, align CUDA/driver versions, tune NCCL and CUDA settings, enable GPU persistence/MIG, pin CPUs and set NUMA affinity, use hugepages and optimized I/O, batch and mixed precision for throughput, cache data, trim images, and add profiling, autoscaling, and reproducibility without sacrificing security.