Scholarlink

With the advancement of deep learning, its application is expanding across various fields. Additionally, as a wide range of Graphics Processing Units (GPUs) become more accessible in the market, researchers face the challenge of selecting optimal GPUs for specific tasks. While GPU architecture has been extensively studied, existing benchmarking efforts often fail to focus on networks that process significantly larger inputs. In this work, we aim to differentiate the performance of various GPUs using a Spatially Varying Bidirectional Reflectance Distribution Function (SVBRDF) extractor network that operates on larger input images (256 \(\,\times \,\) 256).

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

Analyzing GPU Performance in Deep Learning: Insights from Large Image Processing

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Narayan Kandel,
  • Melanie Lambert

摘要

With the advancement of deep learning, its application is expanding across various fields. Additionally, as a wide range of Graphics Processing Units (GPUs) become more accessible in the market, researchers face the challenge of selecting optimal GPUs for specific tasks. While GPU architecture has been extensively studied, existing benchmarking efforts often fail to focus on networks that process significantly larger inputs. In this work, we aim to differentiate the performance of various GPUs using a Spatially Varying Bidirectional Reflectance Distribution Function (SVBRDF) extractor network that operates on larger input images (256 \(\,\times \,\) 256).