DigiNews

Tech Watch by Johan Denoyer

← Back to articles

CUDA Shared Memory Swizzling

Quality: 8/10 Relevance: 9/10

Summary

The article explains how CUDA shared memory bank conflicts can degrade performance and introduces swizzling as a method to avoid these conflicts without padding. It provides the swizzling formula, discusses its properties, and showcases matrix transpose examples comparing bank conflicts, padding, and swizzling, along with performance observations and references.

🚀 Service construit par Johan Denoyer