nvidia-cuda-tileiras

What it is and what it does

nvidia-cuda-tileiras is an assembler package that implements TileIR, a language-agnostic intermediate representation designed for CUDA kernels. It sits between high-level kernel code and low-level CUDA compilation, allowing kernels to be expressed and optimized in a portable form independent of the source language or compilation context.

The package depends on three NVIDIA CUDA runtime components and is distributed as platform-specific wheels for Linux and Windows. It targets developers and researchers working on machine learning, deep learning, and scientific computing who need to work with CUDA kernel compilation at an intermediate abstraction level. The package is in Beta status and actively maintained, though its documentation and public API maturity are not yet established.

Use it for:

• Compile and optimize CUDA kernels through a portable intermediate representation without language-specific tooling
• Integrate kernel compilation into machine learning frameworks that need cross-platform CUDA support
• Research and develop new CUDA optimization techniques using TileIR as a stable compilation target

Worth the install?

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

Provides a portable intermediate representation (TileIR) for CUDA kernels that abstracts away language-specific details, enabling kernel compilation and optimization across different contexts.

Yes, with conditions. Install if you are building CUDA compilation infrastructure or a machine learning framework that needs portable kernel representation; the package is actively maintained and has no known vulnerabilities. Do not install if you need clear documentation or stable public APIs—it is still in Beta. Verify the license terms before use in proprietary projects, as licensing is currently unclear.

Install

nvidia-cuda-tileiras on PyPI

pip install nvidia-cuda-tileiras

uv add nvidia-cuda-tileiras

poetry add nvidia-cuda-tileiras

Installing nvidia-cuda-tileiras

Before you install

Medium install friction due to platform-specific wheels and three NVIDIA CUDA runtime dependencies (nvidia-cuda-nvcc, nvidia-nvvm, nvidia-nvjitlink). Package is actively maintained with a recent release cycle.

License in practice

License status is unclear—no SPDX identifier or raw license text is available in the package metadata. Verify licensing terms before use in proprietary or copyleft-sensitive projects.

Quickstart

pip install nvidia-cuda-tileiras
import tileiras
# Use TileIR assembler for kernel compilation

Requires NVIDIA CUDA runtime libraries (nvidia-cuda-nvcc, nvidia-nvvm, nvidia-nvjitlink) to be installed; platform-specific wheels for Linux (x86_64, aarch64) and Windows (amd64) only.

Verify before relying

• Exact API surface and usage patterns for TileIR intermediate representation
• Whether TileIR is intended for end-user kernel development or primarily for compiler/framework integration
• Documentation availability and maturity level for the assembler interface

Package facts

Evidence: nvidia_cuda_tileiras-13.3.36-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl; nvidia_cuda_tileiras-13.3.36-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl; nvidia_cuda_tileiras-13.3.36-py3-none-win_amd64.whl

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

Tags

More Software Development packages

Further reading