The AI Memory Wall: Why LLMs & AI Accelerators Hit a Bottleneck

Learn more about the Memory Wall here → https://ibm.biz/~NjOWdJHJy More GPU compute doesn't always mean faster AI models. Kaoutar El Maghraoui explains the memory wall, arithmetic intensity, and why LLM inference can become memory-bound or compute-bound. Learn how memory bandwidth, batching, and the roofline model impact AI performance. 00:00 Why Faster Chips Don’t Always Make AI Workloads Faster 00:32 The Memory Wall: The Hidden Bottleneck in Modern Computing 01:00 Compute Growth vs Memory Bandwidth Explained 02:39 Arithmetic Intensity: The Metric That Predicts Performance 03:48 Understan