Storage Implications for the New Generation of AI Applications

Explore the rapidly evolving storage requirements and implications of next-generation AI applications in this comprehensive conference presentation. Examine how the shift from simple data stockpiling to leveraging data for GenAI insights and predictive AI business intelligence is fundamentally changing storage architecture needs. Follow a data-driven analysis to understand where storage performance truly matters - whether in compute nodes or backend systems - and learn to distinguish between the relative importance of bandwidth versus latency in AI workloads. Discover how to evaluate whether the volume and granularity of data accesses are suitable for GPU processing and understand the range of access granularities required by modern AI applications. Analyze recent developments in AI applications and their surprising storage implications, including opportunities for innovative solutions such as NAND architecture adjustments and controller optimizations to reduce power consumption through more efficient data transfers and error checking. Investigate mandatory IOPs optimizations that can benefit current applications and explore new hardware/software codesign approaches that may lead to protocol changes and strategic trade-offs in computing agents and data structures. Learn about novel software interfaces and infrastructure that enable tailored data movement, access, and management specific to each application's needs. Understand how the entire vertical stack - from NAND and controllers to interfaces, buffering, policies, and software abstractions - is being affected by AI application requirements, with every layer of hardware and software experiencing significant changes as organizations build platforms for GenAI insights and predictive AI capabilities.