Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the intricate data flow mechanisms within GPU architectures during Large Language Model training in this comprehensive 49-minute conference presentation. Begin with foundational concepts of neural networks and discover how these complex systems map onto GPU arrays for efficient processing. Examine the current challenges facing GPU topologies and learn how emerging technologies like UEC (Universal Ethernet Consortium) and UALink will transform large-scale GPU array architectures. Master the fundamentals of neural network operations, understand the mapping strategies for distributing neural networks across multiple GPUs, and analyze the critical data flow patterns that occur both within individual GPUs and across interconnected GPU clusters. Gain insights into the pivotal roles that UEC and UALink technologies play in optimizing performance and scalability for massive GPU deployments used in modern AI training workflows.
