#rtxpowered

This article is a part of the AI Decoded series, which shows off new RTX workstation and PC hardware, software, tools, and accelerations while demystifying AI by making the technology more approachable.

AI is fostering innovation and increasing efficiency across sectors, but in order to reach its full potential, the system has to be trained on enormous volumes of high-quality data.

Data scientists are crucial to the preparation of this data, particularly in domain-specific industries where improving AI skills requires specialized, sometimes private data.

NVIDIA revealed that RAPIDS cuDF, a library that makes data manipulation easier for users, speeds up the pandas software library without requiring any code modifications. This is intended to assist data scientists who are facing a growing amount of labor. Pandas is a well-liked, robust, and adaptable Python computer language data analysis and manipulation toolkit. Data scientists may now utilize their favorite code base without sacrificing the speed at which data is processed with RAPIDS cuDF.

Additionally, NVIDIA RTX AI hardware and technology help speed up data processing. Proficient GPUs are among them, providing the computing capacity required to swiftly and effectively boost AI across the board, from data science operations to model training and customization on PCs and workstations.

Python Pandas

Tabular data is the most often used data format; it is arranged in rows and columns. Spreadsheet programs such as Excel may handle smaller datasets; however, modeling pipelines and datasets with tens of millions of rows usually need data frame libraries in Python or other programming languages.

Because of the pandas package, which has an intuitive application programming interface (API), Python is a popular option for data analysis. However, pandas has processing speed and efficiency issues on CPU-only systems as dataset volumes increase. enormous language models need enormous datasets with a lot of text, which the library is infamously bad at handling.

Data scientists are presented with a choice when their data needs exceed pandas’ capabilities: put up with lengthy processing times or make the difficult and expensive decision to migrate to more complicated and expensive technologies that are less user-friendly.

RAPIDS cuDF-Accelerated Preprocessing Pipelines

Data scientists may utilize their favorite code base without compromising processing performance using RAPIDS cuDF.

An open-source collection of the Python packages with GPU acceleration called RAPIDS is intended to enhance data science and analytics workflows. A GPU Data Frame framework called RAPIDS cuDF offers an API for loading, filtering, and modifying data that is similar to pandas.Image credit to Nvidia

Data scientists may take use of strong parallel processing by running their current pandas code on GPUs using RAPIDS cuDF‘s “pandas accelerator mode,” knowing that the code will transition to CPUs as needed. This compatibility offers cutting-edge, dependable performance.

Larger datasets and billions of rows of tabular text data are supported by the most recent version of RAPIDS cuDF. This makes it possible for data scientists to preprocess data for generative AI use cases using pandas code.

NVIDIA RTX-Powered AI Workstations and PCs Improve Data Science

A recent poll indicated that 57% of data scientists use PCs, desktops, or workstations locally.

Significant speedups may be obtained by data scientists beginning with the NVIDIA GeForce RTX 4090 GPU. When compared to conventional CPU-based solutions, cuDF may provide up to 100x greater performance with NVIDIA RTX 6000 Ada Generation GPUs in workstations as datasets expand and processing becomes more memory-intensive.Image credit to Nvidia

With the NVIDIA AI Workbench, data scientists may quickly become proficient with RAPIDS cuDF. Together, data scientists and developers can design, collaborate on, and move AI and data science workloads across GPU systems with our free developer environment manager powered by containers. Several sample projects, like the cuDF AI Workbench project, are available on the NVIDIA GitHub repository to help users get started.

Additionally, cuDF is pre-installed on HP AI Studio, a centralized data science platform intended to assist AI professionals in smoothly migrating their desktop development environment to the cloud. As a result, they may establish, work on projects together, and manage various situations.

Beyond only improving performance, cuDF on RTX-powered AI workstations and PCs has further advantages. It furthermore

Offers fixed-cost local development on strong GPUs that replicates smoothly to on-premises servers or cloud instances, saving time and money.

Enables data scientists to explore, improve, and extract insights from datasets at interactive rates by enabling faster data processing for quicker iterations.

Provides more effective data processing later on in the pipeline for improved model results.

A New Data Science Era

The capacity to handle and analyze large information quickly will become a critical difference as AI and data science continue to advance and allow breakthroughs across sectors. RAPIDS cuDF offers a platform for next-generation data processing, whether it is for creating intricate machine learning models, carrying out intricate statistical analysis, or investigating generative artificial intelligence.

In order to build on this foundation, NVIDIA is supporting the most widely used data frame tools, such as Polaris, one of the fastest-growing Python libraries, which out-of-the-box dramatically speeds up data processing as compared to alternative CPU-only tools.

This month, Polars revealed the availability of the RAPIDS cuDF-powered Polars GPU Engine in open beta. Users of Polars may now increase the already blazingly fast dataframe library’s speed by up to 13 times.