well the binaries don't have icons, and the splash screen is def not robocop (also using directml in 2026???)
March 6, 2026 at 9:58 AM
well the binaries don't have icons, and the splash screen is def not robocop (also using directml in 2026???)
🚀 Unlock blazing-fast neural network inference on AMD hardware!
Discover how to maximize #ONNX Runtime + #DirectML with DirectX 12 for seamless GPU pipelines.
🛠️ Dive into practical setup tips and code samples in our latest guide: gpuopen.com/learn/onnx-d...
Discover how to maximize #ONNX Runtime + #DirectML with DirectX 12 for seamless GPU pipelines.
🛠️ Dive into practical setup tips and code samples in our latest guide: gpuopen.com/learn/onnx-d...
ONNX and DirectML execution provider guide - part 1 - AMD GPUOpen
Learn how to optimize neural network inference on AMD hardware using the ONNX Runtime with the DirectML execution provider and DirectX 12 in the first part of our guide.
gpuopen.com
October 31, 2025 at 8:58 PM
🚀 Unlock blazing-fast neural network inference on AMD hardware!
Discover how to maximize #ONNX Runtime + #DirectML with DirectX 12 for seamless GPU pipelines.
🛠️ Dive into practical setup tips and code samples in our latest guide: gpuopen.com/learn/onnx-d...
Discover how to maximize #ONNX Runtime + #DirectML with DirectX 12 for seamless GPU pipelines.
🛠️ Dive into practical setup tips and code samples in our latest guide: gpuopen.com/learn/onnx-d...
Qualcomm has re-released the withdrawn NPU driver that's needed for both DirectML and ONNX support on Copilot+ hardware. I wonder if my test code will work now…
https://blogs.windows.com/windowsdeveloper/2024/08/29/directml-expands-npu-support-to-copilot-pcs-and-webnn/
https://blogs.windows.com/windowsdeveloper/2024/08/29/directml-expands-npu-support-to-copilot-pcs-and-webnn/
October 29, 2024 at 9:40 PM
Qualcomm has re-released the withdrawn NPU driver that's needed for both DirectML and ONNX support on Copilot+ hardware. I wonder if my test code will work now…
https://blogs.windows.com/windowsdeveloper/2024/08/29/directml-expands-npu-support-to-copilot-pcs-and-webnn/
https://blogs.windows.com/windowsdeveloper/2024/08/29/directml-expands-npu-support-to-copilot-pcs-and-webnn/
⬛ No more black boxes! AMD Radeon GPU Profiler (RGP) now supports AMD RDNA 4 GPUs.
🔥 Profiling of #DirectML apps and compute-based #DX12 and #Vulkan apps
🔥 Improved support for Work Graphs apps
🔥 UI enhancements to the ISA views
More info: gpuopen.com/learn/radeon... (🧵 4/7)
🔥 Profiling of #DirectML apps and compute-based #DX12 and #Vulkan apps
🔥 Improved support for Work Graphs apps
🔥 UI enhancements to the ISA views
More info: gpuopen.com/learn/radeon... (🧵 4/7)
AMD Radeon™ GPU Profiler 2.4 adds support for AMD Radeon™ RX 9000 Series, pure-compute applications, DirectML applications (and more!)
Discover the latest Radeon GPU Profiler v2.4, now supporting Radeon RX 9000 Series GPUs and profiling for pure compute and DirectML applications. Enhance your optimization with improved ISA views and ...
gpuopen.com
March 11, 2025 at 4:12 PM
⬛ No more black boxes! AMD Radeon GPU Profiler (RGP) now supports AMD RDNA 4 GPUs.
🔥 Profiling of #DirectML apps and compute-based #DX12 and #Vulkan apps
🔥 Improved support for Work Graphs apps
🔥 UI enhancements to the ISA views
More info: gpuopen.com/learn/radeon... (🧵 4/7)
🔥 Profiling of #DirectML apps and compute-based #DX12 and #Vulkan apps
🔥 Improved support for Work Graphs apps
🔥 UI enhancements to the ISA views
More info: gpuopen.com/learn/radeon... (🧵 4/7)
directMLは遅すぎて使い物にならないしこんなんじゃあ商品にならないよぉ！！
December 17, 2024 at 10:31 AM
directMLは遅すぎて使い物にならないしこんなんじゃあ商品にならないよぉ！！
*Run trained neural networks on the GPU — your way, your pipeline.*
A pragmatic, three-path series: integrate battle-tested libraries (TensorFlow Lite, ONNX Runtime, PyTorch Mobile, DirectML), compile models through an ML compiler (IREE, TVM, OpenXLA), or […]
[Original post on fosstodon.org]
A pragmatic, three-path series: integrate battle-tested libraries (TensorFlow Lite, ONNX Runtime, PyTorch Mobile, DirectML), compile models through an ML compiler (IREE, TVM, OpenXLA), or […]
[Original post on fosstodon.org]
July 10, 2026 at 12:33 AM
*Run trained neural networks on the GPU — your way, your pipeline.*
A pragmatic, three-path series: integrate battle-tested libraries (TensorFlow Lite, ONNX Runtime, PyTorch Mobile, DirectML), compile models through an ML compiler (IREE, TVM, OpenXLA), or […]
[Original post on fosstodon.org]
A pragmatic, three-path series: integrate battle-tested libraries (TensorFlow Lite, ONNX Runtime, PyTorch Mobile, DirectML), compile models through an ML compiler (IREE, TVM, OpenXLA), or […]
[Original post on fosstodon.org]
Mistral 7B v0.1, MIT, now ONNX-packaged by Microsoft — PyTorch-free across CUDA, DirectML, OpenVINO, ROCm, Ascend. Caveat: no parity benchmarks vs the source weights. Until someone publishes a graph-vs-PyTorch table, I'd keep production on the originals.
Microsoft's ONNX Mistral 7B Hits Modelers.cn: A Portable Base Model for ONNX Runtime
aichina.news
September 18, 2026 at 7:23 PM
Mistral 7B v0.1, MIT, now ONNX-packaged by Microsoft — PyTorch-free across CUDA, DirectML, OpenVINO, ROCm, Ascend. Caveat: no parity benchmarks vs the source weights. Until someone publishes a graph-vs-PyTorch table, I'd keep production on the originals.
그리고 디프싱어 관련으로 환경 설정하실 때 머신 러닝 장치에 CPU로 되어 있는 걸 directml로 바꾼 이후에 단계도 좀 낮은 단계로 설정해야 렉 덜 걸리는 걸로 압니다...
January 3, 2026 at 6:44 AM
그리고 디프싱어 관련으로 환경 설정하실 때 머신 러닝 장치에 CPU로 되어 있는 걸 directml로 바꾼 이후에 단계도 좀 낮은 단계로 설정해야 렉 덜 걸리는 걸로 압니다...
DirectML : Windows ML :: Direct2D : Win2D
I sense a naming pattern
I sense a naming pattern
November 11, 2025 at 7:44 PM
DirectML : Windows ML :: Direct2D : Win2D
I sense a naming pattern
I sense a naming pattern
I was pretty excited about ONNX and DirectML for the privacy potential in the middle of the "shovel everything through OpenAI's servers" maelstrom, and happy to see *some* fraction of my employer's AI river actually stay on-device.
October 26, 2024 at 3:40 PM
I was pretty excited about ONNX and DirectML for the privacy potential in the middle of the "shovel everything through OpenAI's servers" maelstrom, and happy to see *some* fraction of my employer's AI river actually stay on-device.
역시 MS 대단하군요. NVIDIA CUDA 없이도 딥러닝에서 GPU를 쓸 수 있도록 만들어뒀네요. NVIDIA 뿐만 아니라 AMD, Intel, 퀄컴 등 DirectX를 지원하는 GPU는 모두 사용가능합니다. 단, 이름 그대로 DirectML이라서 윈도우에서만 사용가능합니다.
GitHub - 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.
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 ...
github.com
May 23, 2023 at 12:28 AM
역시 MS 대단하군요. NVIDIA CUDA 없이도 딥러닝에서 GPU를 쓸 수 있도록 만들어뒀네요. NVIDIA 뿐만 아니라 AMD, Intel, 퀄컴 등 DirectX를 지원하는 GPU는 모두 사용가능합니다. 단, 이름 그대로 DirectML이라서 윈도우에서만 사용가능합니다.
- AI가 학습 데이터를 기반으로 캐릭터의 세부 사항(머리카락, 화장 등)을 임의로 변경할 수 있으며, 개발자는 색상 보정이나 마스킹 외에 AI의 해석을 세밀하게 제어할 수 있는 수단이 부족
- 화면에 보이는 정보만 활용하는 스크린 공간 효과에 국한되며, 현재로서는 타사 하드웨어와의 호환성(DirectML 등) 계획 없이 엔비디아 하드웨어 전용 기술로 유지되고 있음 (🧵2/2)
- 화면에 보이는 정보만 활용하는 스크린 공간 효과에 국한되며, 현재로서는 타사 하드웨어와의 호환성(DirectML 등) 계획 없이 엔비디아 하드웨어 전용 기술로 유지되고 있음 (🧵2/2)
March 21, 2026 at 11:18 AM
- AI가 학습 데이터를 기반으로 캐릭터의 세부 사항(머리카락, 화장 등)을 임의로 변경할 수 있으며, 개발자는 색상 보정이나 마스킹 외에 AI의 해석을 세밀하게 제어할 수 있는 수단이 부족
- 화면에 보이는 정보만 활용하는 스크린 공간 효과에 국한되며, 현재로서는 타사 하드웨어와의 호환성(DirectML 등) 계획 없이 엔비디아 하드웨어 전용 기술로 유지되고 있음 (🧵2/2)
- 화면에 보이는 정보만 활용하는 스크린 공간 효과에 국한되며, 현재로서는 타사 하드웨어와의 호환성(DirectML 등) 계획 없이 엔비디아 하드웨어 전용 기술로 유지되고 있음 (🧵2/2)
#AIart #Aiイラスト #AI画像
こういうベンチマークが発表されるのは自分としては興味深いです
公式サイト
benchmarks.ul.com/news/announc...
news.mynavi.jp/article/2024...
こういうベンチマークが発表されるのは自分としては興味深いです
公式サイト
benchmarks.ul.com/news/announc...
news.mynavi.jp/article/2024...
「Procyon」に生成AIベンチマーク追加 - ONNX DirectML・NVIDIA TensorRT・Intel OpenVINO対応
UL Solutionsは3月21日（現地時間）、展開中のベンチマークスイート「Procyon」において、生成AI性能の評価に対応する「Procyon AI Image Generation Benchmark」を発表した。3月25日（日本ではおそらく26日）に利用可能になる。
news.mynavi.jp
March 22, 2024 at 5:48 AM
#AIart #Aiイラスト #AI画像
こういうベンチマークが発表されるのは自分としては興味深いです
公式サイト
benchmarks.ul.com/news/announc...
news.mynavi.jp/article/2024...
こういうベンチマークが発表されるのは自分としては興味深いです
公式サイト
benchmarks.ul.com/news/announc...
news.mynavi.jp/article/2024...
Qualcomm seemingly knows OEMs are terrible at releasing drivers because they cleverly titled this blog post about new GPU drivers "for your Microsoft Surface"
www.qualcomm.com/developer/bl...
www.qualcomm.com/developer/bl...
Upgrade to the latest graphics driver for your Microsoft Surface devices powered by Snapdragon X Elite
Upgrade to new Qualcomm Adreno GPU driver for higher performance, DirectML, game fixes, support for new titles, and optimizations for models in Qualcomm AI Hub.
www.qualcomm.com
November 20, 2024 at 7:56 PM
Qualcomm seemingly knows OEMs are terrible at releasing drivers because they cleverly titled this blog post about new GPU drivers "for your Microsoft Surface"
www.qualcomm.com/developer/bl...
www.qualcomm.com/developer/bl...
I'm trying to run an LLM on my laptop locally to generate keywords for images and this sucks. There's like nothing that uses the NPU (either through Onnx, DirectML, or QNN) except for like. Microsoft and Qualcomm models that don't support vision or something. I don't know how to use them.
April 29, 2025 at 3:03 AM
I'm trying to run an LLM on my laptop locally to generate keywords for images and this sucks. There's like nothing that uses the NPU (either through Onnx, DirectML, or QNN) except for like. Microsoft and Qualcomm models that don't support vision or something. I don't know how to use them.
I'd like to interject for a moment. What you're referring to as DirectX is, in fact, Direct3D/DirectDraw/DirectMusic/ DirectPlay/DirectSound/DirectSound3D/DirectInput/DirectCompute/DirectX Raytracing/DirectStorage/DirectML/DXGI/Media Foundation/XACT3/XAudio2/XInput or as I've recently taken to calli
March 23, 2025 at 11:35 PM
I'd like to interject for a moment. What you're referring to as DirectX is, in fact, Direct3D/DirectDraw/DirectMusic/ DirectPlay/DirectSound/DirectSound3D/DirectInput/DirectCompute/DirectX Raytracing/DirectStorage/DirectML/DXGI/Media Foundation/XACT3/XAudio2/XInput or as I've recently taken to calli
祝DirectML環境卒業w
中古とはいえちょっとだけ無理はしたけども
ただ、机を変えた関係で模様替え真っ最中で今週いっぱいはPC触れるか微妙……
中古とはいえちょっとだけ無理はしたけども
ただ、机を変えた関係で模様替え真っ最中で今週いっぱいはPC触れるか微妙……
July 19, 2024 at 1:34 PM
祝DirectML環境卒業w
中古とはいえちょっとだけ無理はしたけども
ただ、机を変えた関係で模様替え真っ最中で今週いっぱいはPC触れるか微妙……
中古とはいえちょっとだけ無理はしたけども
ただ、机を変えた関係で模様替え真っ最中で今週いっぱいはPC触れるか微妙……
Submitted a new issue about torch-directml #DirectML github.com/microsoft/Di...
Multiplication of a float tensor and a bool tensor produces a bool tensor on PyTorch · Issue #686 · microsoft/DirectML
Multiplying a float tensor and a bool tensor on DML produces a bool tensor which is expected to be a float tensor as on CPU. import torch import torch_directml dml = torch_directml.device() t1cpu =...
github.com
February 2, 2025 at 9:45 AM
Submitted a new issue about torch-directml #DirectML github.com/microsoft/Di...
I currently use MPC-HC and MadVR and that requires at least a GTX 1660 to max out the quality sliders for upscaling to 4K. Sadly, none of the Android boxes have the horsepower. DirectML and other heavy upscaling algos with MPV seem to have similar requirements. I'll check ebay!
May 30, 2026 at 3:27 AM
I currently use MPC-HC and MadVR and that requires at least a GTX 1660 to max out the quality sliders for upscaling to 4K. Sadly, none of the Android boxes have the horsepower. DirectML and other heavy upscaling algos with MPV seem to have similar requirements. I'll check ebay!
Yeah, It would be interesting to see the results. I tried DirectML implementation with float16 and it was ~2 times slower
May 20, 2025 at 6:46 AM
Yeah, It would be interesting to see the results. I tried DirectML implementation with float16 and it was ~2 times slower
DirectML Plugin for TensorFlow 2 is here.
#cloud #TensorFlow #directML #wsl #windowsML #PyTorch
https://devblogs.microsoft.com/windowsai/directml-plugin-for-tensorflow-2-is-here/
#cloud #TensorFlow #directML #wsl #windowsML #PyTorch
https://devblogs.microsoft.com/windowsai/directml-plugin-for-tensorflow-2-is-here/
DirectML Plugin for TensorFlow 2 is here - Windows AI Pla...
We are excited to announce the first preview of DirectML ...
devblogs.microsoft.com
November 22, 2024 at 11:49 PM
DirectML Plugin for TensorFlow 2 is here.
#cloud #TensorFlow #directML #wsl #windowsML #PyTorch
https://devblogs.microsoft.com/windowsai/directml-plugin-for-tensorflow-2-is-here/
#cloud #TensorFlow #directML #wsl #windowsML #PyTorch
https://devblogs.microsoft.com/windowsai/directml-plugin-for-tensorflow-2-is-here/
CUDA (NVDAのGPU)がないので、DirectMLのtorchに差し替えて代用した。Direct3D経由になっちゃうので5倍遅いが使えないよりはマシ
Radeon 890M Graphicsでも生成AIはできるよってくらい
絵一枚だけだったら10万円のグラフィックボードなくてもいけるはず
動画とか大量に生成すると非効率極まりないのでそういう人は5090ぽちってください。
Radeon 890M Graphicsでも生成AIはできるよってくらい
絵一枚だけだったら10万円のグラフィックボードなくてもいけるはず
動画とか大量に生成すると非効率極まりないのでそういう人は5090ぽちってください。
April 10, 2025 at 11:43 AM
CUDA (NVDAのGPU)がないので、DirectMLのtorchに差し替えて代用した。Direct3D経由になっちゃうので5倍遅いが使えないよりはマシ
Radeon 890M Graphicsでも生成AIはできるよってくらい
絵一枚だけだったら10万円のグラフィックボードなくてもいけるはず
動画とか大量に生成すると非効率極まりないのでそういう人は5090ぽちってください。
Radeon 890M Graphicsでも生成AIはできるよってくらい
絵一枚だけだったら10万円のグラフィックボードなくてもいけるはず
動画とか大量に生成すると非効率極まりないのでそういう人は5090ぽちってください。
もっと #白フリルドレスお嬢様部 流行れ！
ということで、プロンプトをALTに埋め込みました
一応、1枚目の感じを想定してますが、結構2・3枚目のようなこちらに片手/両手を差し出すポーズが（少なくとも私の環境であるConfyUI+AnimagineV3 3.0+DirectMLでは）出てきます
あとプロンプトの都合、袖が比較的分離しやすいです（4枚目）
特に著作権など主張するつもりはありませんので、改変等はご自由にどうぞ
タグをつけていただけると私のフィードが自動収集してくれるので助かりますｗ
#AIイラスト #AIart
ということで、プロンプトをALTに埋め込みました
一応、1枚目の感じを想定してますが、結構2・3枚目のようなこちらに片手/両手を差し出すポーズが（少なくとも私の環境であるConfyUI+AnimagineV3 3.0+DirectMLでは）出てきます
あとプロンプトの都合、袖が比較的分離しやすいです（4枚目）
特に著作権など主張するつもりはありませんので、改変等はご自由にどうぞ
タグをつけていただけると私のフィードが自動収集してくれるので助かりますｗ
#AIイラスト #AIart
March 4, 2024 at 12:17 PM
もっと #白フリルドレスお嬢様部 流行れ！
ということで、プロンプトをALTに埋め込みました
一応、1枚目の感じを想定してますが、結構2・3枚目のようなこちらに片手/両手を差し出すポーズが（少なくとも私の環境であるConfyUI+AnimagineV3 3.0+DirectMLでは）出てきます
あとプロンプトの都合、袖が比較的分離しやすいです（4枚目）
特に著作権など主張するつもりはありませんので、改変等はご自由にどうぞ
タグをつけていただけると私のフィードが自動収集してくれるので助かりますｗ
#AIイラスト #AIart
ということで、プロンプトをALTに埋め込みました
一応、1枚目の感じを想定してますが、結構2・3枚目のようなこちらに片手/両手を差し出すポーズが（少なくとも私の環境であるConfyUI+AnimagineV3 3.0+DirectMLでは）出てきます
あとプロンプトの都合、袖が比較的分離しやすいです（4枚目）
特に著作権など主張するつもりはありませんので、改変等はご自由にどうぞ
タグをつけていただけると私のフィードが自動収集してくれるので助かりますｗ
#AIイラスト #AIart