Prerequisites#

Before installing TensorRT-RTX, ensure your system meets the following requirements.

Quick Checklist#

✓ NVIDIA GPU (Turing architecture or later) ✓ CUDA Toolkit 12.9 or later installed ✓ Appropriate GPU drivers installed ✓ Python 3.10–3.13 recommended (3.8–3.9 wheels available)

Required Hardware#

NVIDIA RTX GPU with Turing (compute capability 7.5) or later architecture
Supported GPU families: RTX 2000-series (Turing), RTX 3000-series (Ampere), RTX 4000-series (Ada Lovelace), RTX 5000-series (Blackwell)

For a complete list of supported GPUs and precision modes, refer to the Support Matrix.

Required Software#

CUDA Toolkit

NVIDIA CUDA 12.9 or CUDA 13.2 Toolkit

Note

Separate TensorRT-RTX packages are available for CUDA 12.x and CUDA 13.x.

Operating System

Platform	Supported Versions
Windows	Windows 10 x64, Windows 11 x64
Linux	Rocky Linux 8.9, Ubuntu 22.04, Ubuntu 24.04 (x86-64)

NVIDIA Developer Program

You must be a member of the NVIDIA Developer Program to download TensorRT-RTX.

Go to the TensorRT-RTX download page.
Click GET STARTED, then click Download Now.
Select the version of TensorRT-RTX that you are interested in.
Select the checkbox to agree to the license terms.
Click the package you want to install. Your download begins.

Optional Dependencies#

Python Bindings

Python 3.8–3.13 supported - Recommended: Python 3.10–3.13 (fully tested and optimized) - Legacy support: Python 3.8–3.9 (wheels available, but may have limited testing)
pip package manager for installing the TensorRT-RTX Python wheel

PyCUDA

When using TensorRT-RTX with the PyCUDA library in Python, use import pycuda.autoprimaryctx instead of import pycuda.autoinit to avoid device conflicts.

Next Steps#

After verifying prerequisites, proceed to Installing TensorRT-RTX for step-by-step instructions on setting up TensorRT-RTX using:

Windows SDK zip — Extract, add DLLs to your PATH, and optionally install Python bindings
Linux tarball — Extract, set LD_LIBRARY_PATH, and optionally install the Python wheel