Record 096 · Runtimes · 2019
67%19.96.U
DirectML
Microsoft
“A DirectX 12 ML path so AMD, Intel, and NVIDIA laptops can all run models.”
DirectML (and the ONNX EP) is why a Windows box without CUDA can still host a local model. The compatibility GPU API.
Opened vendor-neutral Windows GPU inference.
- DX12 ML API
- ONNX Runtime EP
- Broad GPU coverage
GPU inference on whatever Windows GPU you have.
