AMD GPU inference documentation
MIGraphX, explained for people building real inference systems.
A practical field guide for installing MIGraphX, compiling ONNX models, calling it from C++ and Python, running containers, and designing an AMD backend beside TensorRT.
Input
model.onnx + fp32 tensors
Compile
parse_onnx -> compile(gpu)
Deploy
load .mxr -> run batches
Run a model
Compile ONNX once, load the saved program, and execute inference on an AMD GPU.
Embed in C++
Use parse, compile, save, load, and eval from a native service or video pipeline.
Port from TensorRT
Map CUDA plus TensorRT concepts to HIP plus MIGraphX for an AMD backend.