cuStreamz: A Journey to Develop GPU-Accelerated Streaming Using RAPIDS
, Software Engineeer, NVIDIA
, Software Engineer, NVIDIA
, NVIDIA
Learn how to run streaming applications on GPUs with NVIDIA RAPIDS cuStreamz and how GPUs are the next leap forward in processing high-speed big data stream processing. cuStreamz is the first GPU-accelerated Python-based streaming data processing library built on top of RAPIDS, the GPU-accelerator for data science libraries. cuStreamz aims to accelerate stream processing throughput and lower the total cost of ownership (TCO). We'll deep dive into cuStreamz architecture, covering how we built production-grade streaming features like checkpointing, state management, and accelerated source and sink connectors. Learn about our journey of building real-time streaming analytics pipelines for NVIDIA GeForce NOW cloud gaming using cuStreamz. Early production data pipeline benchmarks have shown a TCO improvement of 4.5x+ as compared to the CPU-based Apache Spark Streaming. We project savings in the magnitude of $100,000s per year, with increased savings expected as more workloads move to GPUs.