https://chkda.github.io/tags/cuda/ Recent content in CUDA on Chhaya's Space https://chkda.github.io/%3Clink%20or%20path%20of%20image%20for%20opengraph,%20twitter-cards%3E https://chkda.github.io/%3Clink%20or%20path%20of%20image%20for%20opengraph,%20twitter-cards%3E Hugo -- 0.163.3 en-us Sun, 05 Jul 2026 00:00:00 +0000 https://chkda.github.io/posts/gpu-internals-speed-run/ Sun, 05 Jul 2026 00:00:00 +0000 https://chkda.github.io/posts/gpu-internals-speed-run/ <p>This article is mostly focussed on CUDA devices. From here onwards whenever I mention GPU know it to be CUDA by NVIDIA.</p> <p>All the notes that have been collected are for my self study. I will put sources below</p> <h1 id="gpu-architecture">GPU Architecture</h1> <p>CUDA stands for Compute Unified Device Architecture.</p> <p><img alt="GPU/CPU architecture" loading="lazy" src="https://chkda.github.io/images/gpu-internal-speed-run/CPU-GPU-Arch.svg"></p> <p><em>Figure adapted from the <a href="https://modal.com/gpu-glossary/">Modal GPU Glossary</a>.</em></p> <p>GPU instructions are carried out by streaming multiprocessors (SMs).</p> <p>An SM in NVIDIA is analogous to a CPU core. SMs execute computation and store state for computation in registers and caches.</p>