跳到正文
NVIDIA Technical Blog· Tanya Lenz·· 26 天前AI 评分12

Scaling NVIDIA TensorRT Performance Optimization Beyond a Single Build Route with the Global Performance Tuner, Part 1

Scaling NVIDIA TensorRT Performance Optimization Beyond a Single Build Route with the Global Performance Tuner, Part 1

AI 导读

部署训练好的神经网络不仅是将算子转换为 GPU 内核,更是一个存在多种合法答案的编译问题。NVIDIA TensorRT 构建引擎时需对图 IR 变换、融合、实现选择、内存放置及量化等相互依赖的决策进行调度。该过程被称为调度(scheduling),是性能优化的核心环节。

来源:NVIDIA Technical Blog · developer.nvidia.com