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

Why watch

Technical deep dive explaining why faster GPU compute doesn't always translate to faster AI model performance. Covers the memory wall phenomenon, how LLM inference can become memory-bound or compute-bound, the roofline model for optimization, and practical techniques like batching to improve throughput. For developers and ML engineers working on AI performance optimization.