Architecture Analysis for ETL Processing - CPU vs GPU

Explore a 19-minute conference talk on GPU acceleration for batch ETL operations. Compare CPU and GPU architectures, including memory subsystems, and analyze database operations like joins, aggregations, and data compression. Discover why these operations are well-suited for GPU acceleration, potentially achieving up to 10x speedup. Examine industry-standard benchmark queries demonstrating full end-to-end SQL query acceleration using GPUs in a prototype query engine, comparing results to existing CPU solutions. Review performance of the same queries at a 3TB scale using the RAPIDS Accelerator for Apache Spark™, enabling GPU acceleration without code changes. Presented by Jason Lowe, Distinguished System Software Engineer at NVIDIA, and Nikolay Sakharnykh, Senior AI Developer Technology Manager at NVIDIA.