Scholarlink

A new way of performing singular value decomposition on a graphic processing unit based on a series of Householder transformations and a Givens rotation matrix principle is proposed and implemented. NVIDIA CUDA programming model and architecture combined with jCUDA library is used for the implementation part. First results have confirmed GPU efficiency claims by previous researchers and demonstrated a great performance in general.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Two-Step Singular Value Decomposition on a Single GPU with CUDA

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Serhii Sukharskyi

摘要

A new way of performing singular value decomposition on a graphic processing unit based on a series of Householder transformations and a Givens rotation matrix principle is proposed and implemented. NVIDIA CUDA programming model and architecture combined with jCUDA library is used for the implementation part. First results have confirmed GPU efficiency claims by previous researchers and demonstrated a great performance in general.