The NVIDIA TensorRT for RTX plugin is a runtime for Unreal Engine’s Neural Network Engine (NNE). Unreal Engine includes multiple NNE runtimes such as CPU and DirectML that can be added to the project via their respective plugins. This plugin adds TensorRT for RTX as an 

additional NNE runtime option. 

TensorRT for RTX is a drop-in replacement for NVIDIA TensorRT in applications targeting NVIDIA RTX GPUs from Turing through Blackwell generations. 

Compared to other NNE runtimes, TensorRT for RTX leverages RTX GPUs across desktops, laptops, and workstations to accelerate real-time, responsive AI workloads including image processing, speech synthesis, and generative AI while maintaining strong performance on consumer-grade devices. It can deliver performance gains of approximately 1.5× to 2× over DirectML. 

The TensorRT for RTX runtime can also be used with existing Unreal Engine features built on NNE, such as Neural Rendering*. 

*Requires minor modifications to Neural Rendering code to add TensorRT for RTX to the list of available runtimes. 

Key Features:

• Evaluate neural networks using Tensor RT-RTX without the need of runtime-specific coding 

• Supports networks with dynamic shapes  

• Inferencing on the GPU, or on GPU aligned with rendering a frame through Render Dependency Graph (RDG) 

• Provides CUDA Graph support optimization 

• Supports reduced-precision data types INT4, INT8, FP4 and FP8 for improved performance at the cost of accuracy 

Requirements:

• Windows x86 

• UE 5.7