NVIDIA TensorRT 如何用 Global Performance Tuner 突破单一构建路线的性能优化（Part 1）
Scaling NVIDIA TensorRT Performance Optimization Beyond a Single Build Route with the Global Performance Tuner, Part 1
NVIDIA TensorRT 构建引擎时需在图变换、算子融合、实现选择、内存布局和量化上做相互依赖的决策，该过程称为调度。Global Performance Tuner 旨在让性能优化不再局限于单一构建路线。原文为 Part 1，未披露具体收益数字。
Deploying a trained neural network involves more than translating each operator into a GPU kernel. It is a compilation problem with many legitimate answers. For...
Deploying a trained neural network involves more than translating each operator into a GPU kernel. It is a compilation problem with many legitimate answers. For example, when NVIDIA TensorRT builds an engine, it makes interdependent decisions about graph-IR transformations, fusion, implementation selection, memory placement, and quantization. This process is called scheduling.
来源：NVIDIA Developer Blog · developer.nvidia.com