Watch a keynote talk from SEBD 2023 (31st Symposium on Advanced Database System) where Carlo Curino from Microsoft discusses how to leverage AI investments in tensor-based computational devices for database analytics acceleration. Learn about Microsoft's research framework that compiles SQL to tensor abstractions compatible with PyTorch/ONNX/TVM runtimes, enabling SQL execution on various GPU/APU platforms. Discover how careful design of tensor-relational-operators and kernel-fusion compilers leads to superior performance and price-performance ratios compared to traditional CPU systems and GPU-native analytics systems. Explore the advantages of blending SQL and ML in novel ways through this approach, implemented in just 20,000 lines of Python code. Gain insights into the future potential of tensor query processing, including performance improvements with newer hardware generations and opportunities for optimization, while understanding the challenges ahead for production deployment.
Overview
Syllabus
Tensor Query Processing: How to ride the AI investment wave for database analytics?
Taught by
IIIA Hub