📌 Introduction Link to heading CUDA Graphs capture a sequence of GPU operations and replay them as a single unit, eliminating the CPU launch overhead for each individual kernel. This is especially beneficial for workloads with small batches or models with many small kernels, where CPU overhead can become a bottleneck. When you use torch.compile(mode="reduce-overhead") or enable…
I’m a performance engineer focused on squeezing every last cycle out of GPU workloads. My interests span high-performance computing, deep learning optimization, and computer architecture — pushing state-of-the-art models to peak performance across speech recognition, machine translation, image classification, and generative AI. This blog is where I document my learning notes in my spare…
Introduction Link to heading The original claim of OpenAI Triton is: We’re releasing Triton 1.0, an open-source Python-like programming language which enables researchers with no CUDA experience to write highly efficient GPU code — most of the time on par with what an expert would be able to produce. The core ideas of Triton : Program GPU with Python : so that the effort to program GPU…