Jawaharlal Nehru Technological University Hyderabad B.Tech III ADVANCED COMPUTER ARCHITECTURE Previous Question Paper (March - 2021 - R18)

The evolution of computer architecture is characterized by a transition from monolithic, single-processor systems to highly parallel, distributed, and specialized computing environments. This progression can be categorized into several distinct generations:

• First Generation (Vacuum Tubes): Based on vacuum tubes, these machines were physically massive, consumed significant power, and relied on machine language programming.
• Second Generation (Transistors): The introduction of transistors replaced vacuum tubes, leading to smaller, more reliable, and faster systems. This era saw the emergence of high-level languages like FORTRAN and COBOL.
• Third Generation (Integrated Circuits): The development of ICs allowed for the miniaturization of components, leading to the birth of minicomputers and the concept of multiprogramming.
• Fourth Generation (VLSI): The advent of Very Large Scale Integration enabled the creation of microprocessors, leading to the personal computer revolution.
• Fifth Generation (Parallelism and AI): Modern architecture focuses on multi-core processors, massive parallelism, heterogeneous computing (CPUs + GPUs/TPUs), and cloud-based distributed architectures to overcome the limitations of sequential processing (the memory wall and power wall).

Parallelism refers to the ability of a system to perform multiple operations simultaneously. The distinction between implicit and explicit parallelism is defined by the responsibility for identifying and managing concurrent tasks:

• Implicit Parallelism: In this model, the hardware or the compiler is responsible for detecting and exploiting parallelism without explicit instructions from the programmer. The compiler analyzes data dependencies and reorders instructions to utilize multiple functional units. Examples include superscalar processors and out-of-order execution engines.
• Explicit Parallelism: Here, the programmer or the language construct explicitly defines which parts of the code can run in parallel. This requires the use of specific parallel programming models or APIs (such as OpenMP, MPI, or CUDA). The programmer is responsible for managing synchronization, data partitioning, and communication between threads or processes.