State of RAPIDS: Bridging the GPU Data Science Ecosystem

Today’s security teams deal with more data than ever before and continue to search for impactful methods to accelerate their log ingest and downstream analytics. In this session, we show how this end-to-end infrastructure is accelerated and augmented using RAPIDS and CLX (Cyber Log Accelerators). We present a use case that shows how raw data can be quickly parsed, enhanced with threat intelligence feeds, and then fed to GPU-based ML techniques for rapid correlation and enrichment. We walk through the architecture, explain how the GPUs are used, present benchmarks and results, and talk some about the larger picture of the value GPU compute brings to cybersecurity workflows.

We'll explore how RAPIDS, the end to end open-source data science accelerator from NVIDIA, and the open-source ecosystem are advancing data science. You'll learn how to start leveraging RAPIDS and the libraries it interacts with for faster performance and easier development on GPUs. We'll focus on the latest engineering work, current benchmarks, new release features, and the future road map. See how RAPIDS works with leading OSS tools like BlazingSQL, Dask, MLFlow, Optuna, Streamz, UCX, Xfeat, and others to deliver high-performance end-to-end data science on GPUs.

We'll step through the process of migrating code from native Python to Numba, and then to a CuPy Raw Kernel (CUDA C++). Basic workflow, best practices, lessons learned, and coding samples will be provided. NVIDIA Nsight Systems profilers will be used to demonstrate how minor optimizations can provide substantial performance benefits to custom developed code. The techniques discussed in this session can be used in any domain.