Multi-GPU Simulation on a Single GPU Machine with MPI

Simulating multi-GPU distributed training on a single-GPU machine using MPI — virtual GPU partitioning, process topology, and gradient synchronization patterns.