High-performance data analytics in Python, at scale.

Distributed

Multi-node data-parallel processing via optimized MPI communication.

Accelerated

Native, out-of-the-box, multi-GPU hardware acceleration via PyTorch.

Scalable

Scale effortlessly beyond single-node RAM limits.

Interoperable

Plug & play compatibility with the Python array ecosystem.

In a nutshell#

Heat builds on PyTorch and mpi4py to process massive arrays - huge collections of images, high-dimensional climate simulation grids, or massive machine learning feature matrices - that exceed the memory and computational limits of a single machine.

Define your data distribution axis via the split parameter, assign hardware using the device attribute, and let Heat orchestrate the parallel computation.

Prototype locally, execute on any cluster.

my_script.py#

import heat as ht

# Distributed random matrix generation
A = ht.random.randn(40000, 10000, split=0, device="gpu")
B = ht.random.randn(10000, 40000, split=1, device="gpu")

# Multi-GPU-accelerated matrix multiplication
C = ht.matmul(A, B)

Run locally or scale across cluster nodes via MPI#

mpirun -np 4 python my_script.py

Getting started#

Quick Install

pip

pip install heat

conda

conda install -c conda-forge heat

HPC & multi-GPU deployments

For Spack, EasyBuild, and containerized setups, refer to our comprehensive deployment guide.

View full installation guide

Latest news#

Jul 2026

New webpage launched

We overhauled our website, let us know what you think!
May 2026

Workshop registration open

Registration is open for our upcoming virtual workshop on high-performance data analytics. Join us for hands-on sessions, expert talks, and live demos of Heat in action.
April 2026

NumFOCUS Affiliation

Heat is officially a NumFOCUS affiliated project! A big milestone for our domain-agnostic library. Check out the announcement details.
March 2026

Version 1.8 Released

Our latest featured update is officially live. Check out the updated repository for full performance metrics.
October 2025

HeatHub ScienceServe Funding

Our project HeatHub has been awarded Helmholtz ScienceServe funding for 2026!