Intel® Distribution for Python*

Achieve near-native code performance with this set of essential packages optimized for high-performance numerical and scientific computing.

High-Performance Python

Intel® Distribution for Python*:

A high-performance Python distribution optimized for Intel® CPUs, GPUs—including current and upcoming Intel® GPU architectures—and accelerators.
Built on open-source Python and enhanced with Intel® performance libraries, it delivers faster analytics, AI, and large-scale scientific workloads across Intel architectures—without code changes or vendor lock-in.
Designed for high performance on Intel® CPUs and GPUs, the distribution supports optimized execution across the full range of Intel® architectures.

For more information, see the system requirements.

Components to develop for Accelerated Compute

Data Parallel Control Library (DPCTL)

This library provides utilities for device selection, allocation of data on devices, tensor data structure, the Python* Array API Standard implementation, and support for the creation of user-defined data-parallel extensions.

Data Parallel Extension for NumPy (DPNP)

This is a drop-in replacement for a subset of NumPy APIs that enable running on Intel CPU and GPUs.

Math Kernel Library (oneMKL) Fourier Transform Functions interface (MKL-FFT)

MKL-based Fast Fourier Transform library for Python. Provides Python interfaces to oneMKL's FFT functions. Esential for signal processing, image analysis and scientific simulations.

Math Kernel Library (OneMKL) Random Number Generation interface (Mkl-Random)

High-performance random number generation for NumPy. Fast and scalable random sampling using Intel's vectorized random engines. Critical for Monte Carlo simulations and ML model initialization.
Math Kernel Library (OneMKL) optimized universal functions "ufuncs" (MKL-UMATH)

Optimized universal functions (ufunc) for element-wise math in NumPy. Exposes Python interface to oneMKL's Vector Math Library. Accelerates operations like exp/log, trigonometric functions.

Who Needs This Product

AI, Machine Learning, and High-Performance Computing (HPC) Developers

For teams building compute-intensive models, pipelines, or simulations that require efficient execution across Intel® CPUs and GPUs. Ideal for training, inference, numerical kernels, and hybrid CPU–GPU workloads.

Data Scientists, Analysts, and Scientific/Engineering Teams

For those working with large datasets, statistical analysis, simulations, or numerical algorithms that benefit from optimized performance on Intel® architectures. Suitable for ETL, data transformations, modeling, and research workflows.

Learners and Students

For individuals exploring Python for AI, data science, or scientific computing using a standards-based environment that runs efficiently on modern Intel® CPUs and GPUs.

What's Included

Package and Environment Managers
Get essential tools for installing, updating, and deleting Python packages and environments.

Data Processing and Modeling Packages
Use these packages in numeric and data science workflows for data collection, ingestion, preprocessing, normalization, transformation, aggregation, and analysis.

Machine Learning Packages
Foundational packages that allow a machine to automatically learn from data without programming it explicitly.

Python Interpreter and Compilers
Use these tools for a versatile interactive experience and to achieve scaled performance.

Advanced Programming Packages
Essential packages that enable fine-grained controls for data management, devices management, concurrency, and parallelism.

Development Packages and Runtimes
Use these runtime packages for enabling performance across Intel-optimized Python packages.

Priority Support
Available through the Intel® oneAPI Base Toolkit.

Benchmarks

Intel-Optimized NumPy & SciPy Linear Algebra Performance

Intel-Optimized NumPy Fast Fourier Transform Performance

Intel-Optimized SciPy Fast Fourier Transform Performance (out-of-place memory placement performance)

Intel-Optimized SciPy Fast Fourier Transform Performance (in-place memory placement performance)