RAPIDS · CUDA-X · Open Source

pandas.
On GPU.
Zero Code Changes.

NVIDIA cuDF is a GPU-accelerated DataFrame library that brings up to 150x faster processing to your existing pandas workflows — with a single line of code.

↗ Get Started on GitHub 📄 Docs

150×

Faster Processing

5GB DuckDB benchmark

30×

Faster on Text

8GB text dataset

Code Changes

Drop-in pandas replacement

9.5M

pandas Users

All can benefit instantly

// The Magic

One Line. Instant GPU Speed.

cudf.pandas intercepts every pandas call and runs it on the GPU first. Falls back to CPU seamlessly for unsupported ops. Your code doesn't know the difference.

📓

Jupyter / IPython

%load_ext cudf.pandas

Magic command — one line at the top of your notebook. All pandas calls in the session are GPU-accelerated.

⚡

Python Script

python -m cudf.pandas script.py

No changes to your script at all. Run with the module flag and get GPU acceleration end-to-end.

🐍

Programmatic

import cudf.pandas; cudf.pandas.install()

Explicitly enable in code when you can't use CLI flags. Works in any Python environment.

cudf_pandas_example.py

# Enable GPU acceleration — just one line!
%load_ext cudf.pandas

# Your existing pandas code unchanged ↓
import pandas as pd

# Read a large parquet file
df = pd.read_parquet("sales_data.parquet")

# Filter, groupby, aggregate — all on GPU
result = (
  df[df["revenue"] > 10000]
  .groupby(["region", "product"])
  .agg({
    "revenue": "sum",
    "units": "mean",
  })
  .sort_values("revenue", ascending=False)
)

# Same pandas API, 150x GPU speed
print(result.head(10))

// Performance

Up to 150× Faster Than pandas

Benchmarks run on NVIDIA A100 vs AMD EPYC CPU. Standard DuckDB database-like operations benchmark.

150×

DuckDB Benchmark (5GB)

Standard analytics ops: joins, groupby, aggregations. Minutes → seconds.

30×

Text Data (8GB)

String operations, file reading, DataFrame merges on large text datasets.

Lines Changed

Zero code changes required. Existing pandas code, third-party libraries — all accelerated.

// Applications

Built for Every Data Workflow

From exploratory data analysis to production ML pipelines, cuDF powers the full data science stack.

📊

Data Science & EDA

Speed up pandas-based exploratory analysis on 2–10 GB datasets with zero refactoring.

💹

Financial Analytics

Real-time stock data processing, risk modeling, and portfolio analytics at GPU speed.

🔄

ETL Pipelines

GPU-accelerated data cleaning, transformation, joins, and aggregations for large datasets.

🤖

ML Feature Engineering

Prepare training data at GPU speed. Integrates with XGBoost, cuML, and PyTorch.

🧬

Genomics & Healthcare

TGen cut 4M-cell analysis from 10 hours to 3 minutes using RAPIDS cuDF.

🌍

Climate Modeling

NASA uses CUDA-X Data Science for air pollution anomaly detection and modeling.

⚡

Apache Spark

RAPIDS Accelerator for Apache Spark — PayPal reduced cloud costs by up to 70%.

📈

Real-Time Dashboards

Power interactive dashboards on large-scale data using cuDF + PyViz libraries.

// Ecosystem

Works With Your Entire Stack

cuDF integrates seamlessly with the Python data science ecosystem — no rewrites needed.

pandas

Drop-in accelerator

Polars

GPU engine backend

Dask

Distributed GPU DFs

Apache Spark

RAPIDS accelerator

XGBoost

GPU gradient boosting

Google Colab

Pre-installed, free GPU

cuML

GPU scikit-learn

Apache Arrow

Columnar memory format

// Premium Domain Available

cuDF.ai

Own the exact-match premium domain for NVIDIA's GPU DataFrame library — the backbone of accelerated data science for 9.5 million pandas users.

✓ Exact-match .ai domain

✓ Massive pandas user base

✓ High SEO potential

✓ Registered on Cloudflare

✓ Clean transfer history

📬 Inquire About Acquisition

[email protected]

Also listed on Afternic · Sedo · Dan.com

pandas.On GPU.Zero Code Changes.

One Line. Instant GPU Speed.

Up to 150× Faster Than pandas

Built for Every Data Workflow

Works With Your Entire Stack

pandas.
On GPU.
Zero Code Changes.