cuda.pathfinder

Type: Technology Tags: NVIDIA, CUDA, Python, Packaging, Libraries, Headers, Tool Discovery Related: CUDA-Python, cuda-bindings, cuda-core, NVCC, libNVVM, libdevice, CUDA-Binary-Utilities, NVIDIA-CUDA Sources: https://nvidia.github.io/cuda-python/cuda-pathfinder/latest/, https://nvidia.github.io/cuda-python/latest/ Last Updated: 2026-04-29

Summary

cuda.pathfinder is a CUDA Python utility package for locating CUDA components installed in a user’s Python environment or CUDA Toolkit environment. It helps Python packages discover NVIDIA dynamic libraries, headers, binary utilities, bitcode libraries, and static libraries without hardcoding local install paths.

Detail

Purpose

CUDA Python packages often need to load native NVIDIA libraries or find headers/tools supplied by pip, Conda, or a system CUDA Toolkit installation. cuda.pathfinder provides a shared discovery layer so packages can locate CUDA resources consistently.

Key capabilities

• Locate CUDA paths from Python environments and CUDA home locations.
• Load supported NVIDIA dynamic libraries.
• Find CUDA Toolkit headers.
• Locate binary utilities such as compiler or binary-inspection tools.
• Find bitcode libraries and static libraries used by compiler workflows.
• Surface structured errors when required CUDA components are missing or unavailable.

NVIDIA context

cuda.pathfinder is not a user-facing GPU programming model like cuda-core or cuTile. Its strategic value is packaging and deployment: it makes Python CUDA packages more robust when CUDA libraries come from a mix of Python wheels, Conda packages, and system installations.

Connections

• CUDA-Python - umbrella that lists cuda.pathfinder as a current component.
• cuda-bindings and cuda-core - Python CUDA packages that may need consistent CUDA component discovery.
• NVCC and CUDA-Binary-Utilities - examples of tools that Python workflows may need to locate.
• libNVVM and libdevice - compiler components/bitcode libraries that can be discovered by package tooling.
• NVIDIA-CUDA - parent platform whose installed components are being discovered.

Source Excerpts

• NVIDIA describes cuda.pathfinder as utilities for locating CUDA components.
• The API reference lists helpers for dynamic libraries, headers, binary utilities, bitcode libraries, and static libraries.