Cengiz, M Theophane (2022) Optimisation de code sur carte graphique PRE - Research Project, ENSTA.



This study aims at establishing how code optimizations realized for a given Graphics Processing Unit (GPU) hardware are affected when replacing said hardware with another. A GEMV algorithm is developped under 3 iterations, until it achieves similar performances to NVidia’s dense linear algebra library (cuBLAS). Afterward, the code is translated for AMD architectures, and further tested to discover new optimizations rules that apply to both NVidia and AMD material. Final code is tested on the ARCANE platform [1], to witness how performances are affected.

Item Type:Thesis (PRE - Research Project)
Uncontrolled Keywords:Parallel computing, Linear Algebra, Graphics Card, Code optimization, Benchmark
Subjects:Information and Communication Sciences and Technologies
Mathematics and Applications
ID Code:9206
Deposited By:Theophane Cengiz
Deposited On:05 juin 2023 10:50
Dernière modification:05 juin 2023 10:50

Repository Staff Only: item control page