Join processing is one of the most performance-critical components of query execution in relational database systems. While classical join algorithms have been extensively studied, their evaluation has traditionally been guided by asymptotic complexity and disk-oriented cost models. Recent advances in processor, memory, and storage architectures—characterized by multi-core processors and deep cache hierarchies—have introduced new dimensions that reshape the performance behavior of join strategies. A comparative, hardware-aware analysis is presented to examine how nested loop, index nested loop, hash join, sort-merge join, and accelerator-based joins interact with underlying hardware properties. The study synthesizes insights from foundational and recent research to analyze join sensitivity to cache locality, memory bandwidth, NUMA effects, parallel scalability, accelerator utilization, and storage characteristics. A structured comparison and hardware-layer mapping illustrate why theoretically efficient join strategies may underperform on contemporary platforms. The analysis indicates that join performance is now dominated more by architectural constraints than by asymptotic algorithmic complexity, and that no single join strategy is universally optimal across heterogeneous hardware environments. The paper concludes by identifying key research challenges as hardware-aware cost modeling, adaptive join execution, and cross-layer optimization - that are critical for the design of efficient and scalable database systems.