NNE GPU Inference

ClaudusBridge integrates with Unreal Engine's Neural Network Engine (NNE) to enable GPU-accelerated inference via DirectML Tensor Cores. Load ONNX models and run inference directly on your GPU.

---

Overview

NNE (Neural Network Engine) is UE5's built-in framework for neural network inference. ClaudusBridge exposes it through MCP tools, allowing AI clients to:

• Load ONNX models onto the GPU
• Run inference using DirectML (Tensor Cores on NVIDIA/AMD GPUs)
• Fall back to CPU inference when GPU is unavailable
• Check available runtimes and hardware capabilities

---

Available Tools

Tool	Description
load_nne_model	Load an ONNX model for GPU inference
get_nne_status	Check NNE initialization and model status
list_nne_runtimes	Discover available inference backends

---

Check Available Runtimes

curl -X POST http://localhost:3000/mcp \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
    "name":"list_nne_runtimes","arguments":{}}}'

Common runtimes:

• NNERuntimeORTDml — GPU inference via DirectML (recommended)
• NNERuntimeORTCpu — CPU fallback

---

Check NNE Status

curl -X POST http://localhost:3000/mcp \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
    "name":"get_nne_status","arguments":{}}}'

Returns:

• Initialization state
• Available runtimes
• Loaded model info (if any)
• GPU capabilities

---

Load an ONNX Model

curl -X POST http://localhost:3000/mcp \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
    "name":"load_nne_model","arguments":{
      "model_path":"/Game/Models/MyModel.onnx",
      "runtime":"NNERuntimeORTDml"}}}'

The model loads onto the GPU via DirectML. If the specified runtime is unavailable, it falls back to CPU.

---

Hardware Requirements

Feature	Requirement
GPU Inference	DirectX 12 compatible GPU
Tensor Cores	NVIDIA RTX series (optional, improves performance)
CPU Fallback	Any x64 CPU
UE Version	Unreal Engine 5.7 (NNE module)

---

Use Cases

• Scene segmentation — Classify viewport regions for navigation
• Object detection — Identify objects in the viewport
• Style transfer — Apply artistic styles to materials
• Custom AI models — Run any ONNX model inside the editor

---

Notes

• NNE is a UE5 experimental feature — API may change between engine versions
• DirectML provides broad GPU compatibility (NVIDIA + AMD + Intel)
• Model loading is asynchronous — check status after loading
• Large models may take several seconds to load onto GPU memory