Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the critical but often overlooked role of storage devices in AI data center infrastructure through this 47-minute conference presentation from SNIA SDC 2025. Examine how the transformational launch of GPT-4 has accelerated the race to build AI data centers for large-scale training and inference, with particular focus on storage components beyond the well-known GPUs and high-bandwidth memory. Delve into the AI processing pipeline within data centers to understand how storage devices such as SSDs function in both compute and storage nodes. Analyze the specific characteristics of AI workloads to derive targeted requirements for flash storage devices and controllers. Discover why storage remains the slowest component in the AI data path compared to compute and memory, and learn how training and inference workloads impose different requirements on storage devices. Understand how AI is driving the adoption of next-generation interfaces for storage devices in data center environments. Investigate essential optimizations needed in NAND media, SSD controllers, host interfaces, and form factors to meet the performance requirements of AI workloads, with insights from Marvell's Erich Haratsch on the technical considerations for flash storage in AI infrastructure.
