Scholarlink

We present a new out-of-core method for multi-GPU path tracing of large scenes based on memory access analysis. Our approach allows us to render massive scenes efficiently on a computer system with multiple GPUs and less total memory of all graphics cards than the total size of the scene. The scene is partitioned among GPU memories and the main CPU memory based on a method that operates at the memory management level. Specific parts of the scene are either replicated or distributed in the memory of GPUs, and the rest is located in the main CPU memory.

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

Multi-GPU Accelerated Rendering of Massive Scenes with Out-of-Core Support for CPU Memory

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Milan Jaros,
  • Lubomir Riha,
  • Petr Strakos,
  • Tomas Kozubek

摘要

We present a new out-of-core method for multi-GPU path tracing of large scenes based on memory access analysis. Our approach allows us to render massive scenes efficiently on a computer system with multiple GPUs and less total memory of all graphics cards than the total size of the scene. The scene is partitioned among GPU memories and the main CPU memory based on a method that operates at the memory management level. Specific parts of the scene are either replicated or distributed in the memory of GPUs, and the rest is located in the main CPU memory.