nvidia-cuda-nvdisasm

What it is and what it does

nvidia-cuda-nvdisasm is a Python package that wraps NVIDIA's nvdisasm tool, allowing you to disassemble CUDA cubin files (compiled GPU binaries) into human-readable CUDA assembly code. It accepts cubin files as input and outputs the corresponding assembly representation, making it possible to inspect what code the NVIDIA compiler actually generated for your GPU kernels.

The package is distributed as platform-specific wheels for x86_64 Linux, aarch64 Linux, and Windows, with no Python runtime dependencies. It targets developers and researchers working with CUDA who need to inspect, debug, or analyze compiled GPU code at the assembly level. The package is actively maintained and supports Python 3.5 through 3.11.

Use it for:

• Analyze compiled CUDA kernel binaries to understand GPU code generation and optimization behavior
• Debug GPU kernel performance by inspecting the assembly-level instructions generated by the NVIDIA compiler
• Reverse-engineer or audit CUDA applications by disassembling their compiled cubin objects
• Validate compiler output and verify that optimization passes are applied as expected

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

nvdisasm disassembles NVIDIA CUDA cubin files into human-readable CUDA assembly code, extracting and presenting the compiled GPU kernel information.

Yes, if you need to disassemble CUDA cubin files for kernel analysis or debugging. Install friction is moderate due to platform-specific wheels and likely system-level CUDA dependencies, but the package is actively maintained and has no Python-level dependency complications. Verify beforehand that your system has the necessary CUDA runtime libraries and that your platform (x86_64 Linux, aarch64 Linux, or Windows) is supported.

Install

nvidia-cuda-nvdisasm on PyPI

pip install nvidia-cuda-nvdisasm

uv add nvidia-cuda-nvdisasm

poetry add nvidia-cuda-nvdisasm

Installing nvidia-cuda-nvdisasm

Before you install

Medium install friction due to platform-specific wheel distribution (x86_64 Linux, aarch64 Linux, Windows). Active maintenance with a recent release within 46 days. No runtime dependencies to manage.

Quickstart

pip install nvidia-cuda-nvdisasm==13.3.73

import nvidia_cuda_nvdisasm
# Use the package to disassemble cubin files

Requires a compatible platform (x86_64 Linux, aarch64 Linux, or Windows); cubin input files must be available; may require NVIDIA CUDA Toolkit or runtime libraries to be installed on the system.

Verify before relying

• Whether this is a wrapper around the standalone NVIDIA nvdisasm binary or a pure Python implementation
• Whether CUDA Toolkit or other system libraries must be pre-installed for this package to function
• What the actual output format and verbosity options are beyond 'human readable format'

Package facts

Evidence: nvidia_cuda_nvdisasm-13.3.73-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl; nvidia_cuda_nvdisasm-13.3.73-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl; nvidia_cuda_nvdisasm-13.3.73-py3-none-win_amd64.whl

Keywords: cuda, nvidia, runtime, machine learning, deep learning

Tags

More Software Development packages

Further reading