Timothy Morano
Could 31, 2025 05:45
NVIDIA’s RAPIDS introduces zero-code acceleration for machine studying, boosts IO efficiency, and helps out-of-core XGBoost coaching, streamlining information science workflows.
NVIDIA has unveiled vital developments in its RAPIDS software program suite, specializing in machine studying acceleration and efficiency enhancements. Based on NVIDIA, the most recent updates introduce zero-code-change acceleration for Python machine studying, substantial IO efficiency enhancements, and help for out-of-core XGBoost coaching.
Zero-Code-Change Acceleration
The brand new capabilities of NVIDIA’s cuML now permit information scientists to leverage zero-code-change acceleration of their workflows. This performance is especially useful for customers of fashionable libraries equivalent to scikit-learn, UMAP, and hdbscan. By using NVIDIA GPUs, information scientists can obtain efficiency beneficial properties of 5-175x with out altering their present codebases.
IO Efficiency Enhancements
RAPIDS’ cuDF has obtained vital efficiency boosts, significantly for cloud-based information processing duties. The mixing of NVIDIA KvikIO permits sooner studying of Parquet information from cloud storage options like Amazon S3, attaining a threefold enchancment in learn speeds. Moreover, the hardware-based decompression engine in NVIDIA’s Blackwell structure facilitates sooner information processing by lowering latency and growing throughput.
Out-of-Core XGBoost Coaching
In collaboration with the DMLC group, RAPIDS has optimized XGBoost for giant datasets, permitting for environment friendly coaching on information exceeding in-memory limits. This improvement is very advantageous for programs using NVIDIA’s GH200 Grace Hopper and GB200 Grace Blackwell, enabling them to deal with datasets over 1 TB effectively.
Usability and Platform Updates
RAPIDS has additionally enhanced usability with options like international configuration settings and GPU-aware profiling for the Polars engine, making it simpler for customers to optimize their information science workflows. Moreover, help for NVIDIA Blackwell-architecture GPUs and enhancements in Conda package deal administration have been launched, broadening the platform’s accessibility and ease of use.
These updates, showcased at NVIDIA GTC 2025, underline NVIDIA’s dedication to advancing information science know-how and streamlining machine studying processes. For extra detailed data on these developments, go to the NVIDIA weblog.
Picture supply: Shutterstock