https://sankethbk.github.io/blog/tags/memory-coalescing/Recent content in Memory-Coalescing on Sanketh's BlogHugo -- 0.166.0en-usSun, 30 Aug 2026 00:00:00 +0000https://sankethbk.github.io/blog/posts/gpu/2026-08-30-3-cuda-memory-coalescing/Sun, 30 Aug 2026 00:00:00 +0000https://sankethbk.github.io/blog/posts/gpu/2026-08-30-3-cuda-memory-coalescing/<h1 id="cuda-memory-coalescing-why-access-patterns-matter">CUDA Memory Coalescing: Why Access Patterns Matter</h1> <p>The <a href="https://sankethbk.github.io/blog/gpu/2026-08-30-2-cuda-programming-model">previous note</a> showed how to write and launch a CUDA kernel. This note answers the first performance question: why does the same kernel sometimes run 5× or 10× slower just because of how threads read memory?</p> <p>The answer is <strong>memory coalescing</strong>. Because GPU memory is optimized for bandwidth over latency, the hardware rewards access patterns where threads in a warp read or write contiguous addresses together. If they do not, the GPU wastes bandwidth and cycles fetching data that most threads ignore.</p>