This thesis analyzes computational performance comparing matrix-vector (MxV) and matrix-matrix (MxM) operations across CPU and GPU architectures. The study demonstrates that matrix-matrix operations consistently outperform matrix-vector operations due to superior operational intensity and better cache utilization.
Note
Block I: Synthetic performance benchmarks (CPU vs GPU)
Block II: Real-world validation using wave equation solvers
Tip
This repository functions as an interactive computational document - every benchmark and experiment can be reproduced on your hardware through an executable menu system.
Launch the main script (main.jl) to reproduce everything in your local machine.
Pedro Rodríguez Jiménez
Aerospace Engineering Bachelor's Thesis
Technical University of Madrid (UPM)
September 2025
Supervisors:
Dr. Juan Antonio Hernández Ramos
Dr. Javier Luis González Monge