System Requirements
Supported Unreal Engine version:
- 5.3
- 5.4
- 5.5
However, when using CUDA or TensorRT, this plugin cannot be used at the same time as the official "NNERuntimeORT" plugin.
Supported Platforms
| Platform | Development | Target Build |
|---|---|---|
| Windows 10 64bit | ✅ | ✅ |
| Ubuntu 18.04.6 Desktop 64bit | ⚠️(Experimental) | ⚠️(Experimental) |
| Android | ⚠️(Experimental) |
Tested only on Windows
Versions for UE5.3 and later have only been tested on Windows.
They have not been fully tested on Ubuntu and Android and we cannot guarantee that they will work properly. Therefore, they are marked as "Experimental".
Model format for Android
To use on Android, DNN models need to be converted to ORT format. See the official document for the details.
Known issue
There are some DNN models which cannot be loaded on Android.
Supported hardware acceleration
| Platform | Default CPU | GPU DirectML | GPU CUDA | GPU TensorRT | NNAPI |
|---|---|---|---|---|---|
| Windows 10 64bit | ✅ | ✅ | ✅ | ✅ | |
| Ubuntu 18.04.6 Desktop 64bit | ✅ | ✅ | ✅ | ||
| Android | ✅ | (Not tested yet) |
- To use GPU acceleration with DirectML, a DirectX 12 capable GPU is required.
- To use GPU acceleration with CUDA and TensorRT,
- a supported NVIDIA GPU is required.
- you need to disable "NNERuntimeORT" which is enabled by default sinve UE5.5.
- the following versions of CUDA, cuDNN, and TensorRT are required to be installed.
CUDA, cuDNN, TensorRT versions
| Other than RTX30** series | RTX30** series | |
|---|---|---|
| CUDA | 11.0.3 | 11.0.3 |
| cuDNN | 8.0.2 (July 24th, 2020), for CUDA 11.0 | 8.0.5 (November 9th, 2020), for CUDA 11.0 |
| TensorRT | 7.1.3.4 for CUDA 11.0 | 7.2.2.3 for CUDA 11.0 |
For RTX2080Ti users
If you are using RTX2080Ti, you might need to use cuDNN v8.0.5. Please try various combinations.
| CUDA | 11.4.2 for Linux x86_64 Ubuntu 18.04 |
| cuDNN | 8.2.4 (September 2nd, 2021), for CUDA 11.4, Linux x86_64 |
| TensorRT | 8.2.3.0 (8.2 GA Update 2) for Linux x86_64, CUDA 11.0-11.5 |
Supported operators for TensorRT
On Linux, DNN models which contain unsupported operators cannot be loaded when TensorRT is enabled.
See the official document for supported operators.
(NNEngine uses TensorRT 8.2 as backend on Linux)