Share to: share facebook share twitter share wa share telegram print page

DirectML

DirectML
開発元 マイクロソフト
初版 2019年 (6年前) (2019)
最新版
1.15.4 / 2024年10月28日 (11か月前) (2024-10-28)[1]
リポジトリ github.com/microsoft/DirectML
対応OS Windows 10 バージョン 1903以降
後継 Windows ML
サポート状況 開発終了
ライセンス MIT License
公式サイト microsoft.github.io/DirectML/
テンプレートを表示

DirectML(Direct Machine Learning)は、マイクロソフトが提供していた、Microsoft Windows上でGPUもしくはNPUを用いてニューラルネットワークディープラーニング)を実行するためのC++の低レベルAPI[2]2025年7月19日にDirectMLは開発終了し[3][4]、代わりにマイクロソフトは2025年5月にWindows MLを発表し、ONNX Runtimeが存続する[5]Windows 10 バージョン 1903以降とWindows 11に標準搭載されていた。

学習結果の推論だけでなく、学習の用途にも利用可能である。ネイティブWindowsだけでなく、Windows上のWindows Subsystem for Linuxからも利用可能である。[6]

マイクロソフトは何度も類似のニューラルネットワークのライブラリを発表して、廃止している。

  1. 2018年に発表して、開発終了したWindows Machine Learning。名前空間はWindows.AI.MachineLearning。[7]
  2. 2019年に発表して、2025年に開発終了したDirectML。名前空間はMicrosoft.AI.DirectML。
  3. 2025年に発表した、Windows ML。名前空間はMicrosoft.Windows.AI.MachineLearning。

対応ハードウェア

GPUの場合は、2015年に発表されたMicrosoft DirectX 12に対応している必要がある。事実上、現在使われているほとんどのGPUが対応している。[8]

  • AMD GCN 第1世代 (Radeon HD 7000 シリーズ) 以降 (2012年)
  • Intel Haswell (第4世代Intel Core) HD Integrated Graphics 以降 (2013年)
  • NVIDIA Kepler マイクロアーキテクチャ (GeForce 600 シリーズ) 以降 (2012年)
  • Qualcomm Adreno 600 以降

NPUは以下のものが対応している。

対応ライブラリ

直接DirectMLを呼び出すことも可能であるが、以下のライブラリが対応している。

  • ONNX Runtime[11]
  • PyTorch。torch-directmlパッケージをインストールする。[12]
  • Web Neural Network APIを実装したWindows上のウェブブラウザ。これ自体もONNX Runtime Webから利用可能。[13]

以下のものは開発が止まっている。

  • TensorFlow。tensorflow-directml-pluginもしくはtensorflow-directmlパッケージをインストールする。[14][15][16]

出典

  1. ^ Microsoft.AI.DirectML”. nuget.org. 2025年10月2日閲覧。
  2. ^ DirectML の概要” (2025年2月10日). 2025年3月1日閲覧。
  3. ^ Update README to indicate DML is in maintenance mode by jstoecker · Pull Request #710 · microsoft/DirectML”. 2025年10月1日閲覧。
  4. ^ Is DML being deprecated? · Issue #23783 · microsoft/onnxruntime”. 2025年10月1日閲覧。
  5. ^ Introducing Windows ML: The future of machine learning development on Windows”. Windows Developer Blog. 2025年10月1日閲覧。
  6. ^ WSL での GPU アクセラレーション” (2024年3月15日). 2025年3月1日閲覧。
  7. ^ Introduction to Windows Machine Learning”. 2025年10月1日閲覧。
  8. ^ microsoft/DirectML: DirectML is a high-performance, hardware-accelerated DirectX 12 library for machine learning. DirectML provides GPU acceleration for common machine learning tasks across a broad range of supported hardware and drivers, including all DirectX 12-capable GPUs from vendors such as AMD, Intel, NVIDIA, and Qualcomm.”. 2025年3月1日閲覧。
  9. ^ Pralle, Chad. “Introducing Neural Processor Unit (NPU) support in DirectML (developer preview) - DirectX Developer Blog”. DirectX Developer Blog. 2025年3月1日閲覧。
  10. ^ Blog, Windows Developer. “DirectML expands NPU support to Copilot+ PCs and WebNN”. Windows Developer Blog. 2025年3月1日閲覧。
  11. ^ Windows - DirectML”. onnxruntime. 2025年3月1日閲覧。
  12. ^ DirectML/PyTorch at master · microsoft/DirectML”. 2025年3月1日閲覧。
  13. ^ Web”. onnxruntime. 2025年3月1日閲覧。
  14. ^ TensorFlow 2 用 DirectML プラグイン” (2025年2月10日). 2025年3月1日閲覧。
  15. ^ microsoft/tensorflow-directml-plugin: DirectML PluggableDevice plugin for TensorFlow 2”. 2025年3月1日閲覧。
  16. ^ microsoft/tensorflow-directml: Fork of TensorFlow accelerated by DirectML”. 2025年3月1日閲覧。

外部リンク

Prefix: a b c d e f g h i j k l m n o p q r s t u v w x y z 0 1 2 3 4 5 6 7 8 9

Portal di Ensiklopedia Dunia

Kembali kehalaman sebelumnya