data parallelism

A practical, bash-first guide to building a reliable multi‑GPU AI setup on Linux: verify hardware/topology, power and cooling; install dev tools and NVIDIA/ROCm drivers; create a Python venv with GPU PyTorch; configure NCCL/RCCL; and run PyTorch DDP with torchrun on single or multi‑node. Includes copy‑paste commands, performance tuning (batch size, mixed precision, data I/O, topology), troubleshooting tips, and optional Docker workflow.