Topic

From Compute Optimization to “Good Code” in the Eyes of the GPU

In the AI era, GPU computing power is as precious as time itself — every second counts. How can we make the most of this valuable resource? This talk takes NVIDIA GPUs and CUDA as examples to introduce the unique characteristics of the GPU software stack and GPU programming. It explores key ideas, tools, and methodologies for optimizing GPU performance, and shares insights on how to design code that GPUs “love” — efficient, parallel-friendly, and fully exploiting the underlying hardware capabilities.