<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/windows-ml-adds-mlir-based-local-ai-inference-with-dx-cgc-k4riv0sp3" -->

---
title: Windows ML Adds MLIR-Based Local AI Inference with DX CGC
description: Windows ML 2.9.42 Experimental adds a new MLIR-based backend called DirectX Compute Graph Compiler (DX CGC), built with AMD, Intel, NVIDIA, and Qualcomm. DX...
canonical: https://daily.dev/posts/windows-ml-adds-mlir-based-local-ai-inference-with-dx-cgc-k4riv0sp3
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Windows ML Adds MLIR-Based Local AI Inference with DX CGC | daily.dev
og:description: Windows ML 2.9.42 Experimental adds a new MLIR-based backend called DirectX Compute Graph Compiler (DX CGC), built with AMD, Intel, NVIDIA, and Qualcomm. DX...
og:url: https://daily.dev/posts/windows-ml-adds-mlir-based-local-ai-inference-with-dx-cgc-k4riv0sp3
og:image: https://api.daily.dev/og/posts/k4riv0sp3.png
og:image:alt: Windows ML Adds MLIR-Based Local AI Inference with DX CGC
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Windows ML Adds MLIR-Based Local AI Inference with DX CGC

**[DevBlogs](https://daily.dev/sources/devblogs)** · 11 min read · 0 upvotes · 0 comments

## Summary

Windows ML 2.9.42 Experimental adds a new MLIR-based backend called DirectX Compute Graph Compiler (DX CGC), built with AMD, Intel, NVIDIA, and Qualcomm. DX CGC compiles models to a portable Compute Graph IR (CGIR) and runs them through Direct3D 12, with shader-based fallbacks when vendor drivers don't support it yet. This release supports GPU workloads with the NVIDIA GeForce 621.24 driver, while CPU and NPU still run through ONNX Runtime. Task APIs now accept GGUF and Hugging Face SafeTensors directly, letting you run models like Qwen3.8-27B locally as the backend for GitHub Copilot CLI with no per-token cost.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://devblogs.microsoft.com/foundry-on-windows/winml-cgc>

## Questions this post answers

### What is DirectX Compute Graph Compiler (DX CGC) in Windows ML?

DX CGC is an MLIR-based compiler and inference engine for graph-based computation in Windows ML, introduced with Windows ML 2.9.42 Experimental. It compiles a portable Compute Graph IR (CGIR) representation of a model down to platform-specific code and executes it through Direct3D 12, using compiler backends built with AMD, Intel, NVIDIA, and Qualcomm. It falls back to shader-based execution when vendor drivers don't yet support its interfaces.

_Developers evaluating new local-inference backends can track releases like this one on daily.dev._

### Can Windows ML run GGUF and Hugging Face SafeTensors models directly without converting to ONNX?

Yes. The Windows ML Task APIs accept GGUF and Hugging Face SafeTensors model files directly as input alongside ONNX, so no conversion or export step is required. With Windows ML 2.9.42 Experimental, SafeTensors models run across text generation, speech recognition, and decision tasks, while GGUF files are supported for text generation, including via a llama.cpp loader with a CUDA GGML backend for GPU acceleration.

_Teams choosing between model formats for local AI apps can follow this kind of compatibility news on daily.dev._

### Which NVIDIA driver is needed for Windows ML DX CGC GPU acceleration?

Vendor-specific GPU acceleration for DX CGC in this initial release requires the NVIDIA GeForce 621.24 driver, or the equivalent 621.24 driver for RTX Spark. Support for additional GPU vendors is planned for future releases; meanwhile CPU and NPU workloads run through the Runtime APIs via ONNX Runtime-based execution instead of DX CGC.

_Developers setting up local GPU inference pipelines can keep driver and compatibility updates on daily.dev._

## Similar posts on daily.dev

- [MiniDXNN v0.4.0: Interactive neural texture compression on DirectX 12](https://daily.dev/posts/minidxnn-v0-4-0-interactive-neural-texture-compression-on-directx-12-51rp9ezlh) · AMD GPUOpen · 0 upvotes · 0 comments
- [Microsoft Updates DirectX Shader Compiler With Improved Vulkan Driver Interoperability](https://daily.dev/posts/microsoft-updates-directx-shader-compiler-with-improved-vulkan-driver-interoperability-ztxfrpbxj) · Phoronix · 0 upvotes · 0 comments
- [NVIDIA Levels Up Local AI Agents Across RTX PCs and DGX Spark](https://daily.dev/posts/nvidia-levels-up-local-ai-agents-across-rtx-pcs-and-dgx-spark-du5cveehq) · NVIDIA · 0 upvotes · 0 comments

---

Tags: [#local-ai](https://daily.dev/tags/local-ai)

[View this post on daily.dev](https://daily.dev/posts/windows-ml-adds-mlir-based-local-ai-inference-with-dx-cgc-k4riv0sp3)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Windows ML Adds MLIR-Based Local AI Inference with DX CGC","url":"https://daily.dev/posts/windows-ml-adds-mlir-based-local-ai-inference-with-dx-cgc-k4riv0sp3","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/windows-ml-adds-mlir-based-local-ai-inference-with-dx-cgc-k4riv0sp3"},"datePublished":"2026-11-22T08:49:34.711Z","dateModified":"2026-11-22T08:51:37.193Z","description":"Windows ML 2.9.42 Experimental adds a new MLIR-based backend called DirectX Compute Graph Compiler (DX CGC), built with AMD, Intel, NVIDIA, and Qualcomm. DX...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4a5b534ec8843e4ddfab8172b2a5c3d4?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4a5b534ec8843e4ddfab8172b2a5c3d4?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"DevBlogs","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"DevBlogs","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/8f68b453325f482ebeb73fb780092713","url":"https://daily.dev/sources/devblogs"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/windows-ml-adds-mlir-based-local-ai-inference-with-dx-cgc-k4riv0sp3","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"local-ai","timeRequired":"PT11M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"DevBlogs","item":"https://daily.dev/sources/devblogs"},{"@type":"ListItem","position":3,"name":"Windows ML Adds MLIR-Based Local AI Inference with DX CGC"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/windows-ml-adds-mlir-based-local-ai-inference-with-dx-cgc-k4riv0sp3#faq","mainEntity":[{"@type":"Question","name":"What is DirectX Compute Graph Compiler (DX CGC) in Windows ML?","acceptedAnswer":{"@type":"Answer","text":"DX CGC is an MLIR-based compiler and inference engine for graph-based computation in Windows ML, introduced with Windows ML 2.9.42 Experimental. It compiles a portable Compute Graph IR (CGIR) representation of a model down to platform-specific code and executes it through Direct3D 12, using compiler backends built with AMD, Intel, NVIDIA, and Qualcomm. It falls back to shader-based execution when vendor drivers don't yet support its interfaces. Developers evaluating new local-inference backends can track releases like this one on daily.dev."}},{"@type":"Question","name":"Can Windows ML run GGUF and Hugging Face SafeTensors models directly without converting to ONNX?","acceptedAnswer":{"@type":"Answer","text":"Yes. The Windows ML Task APIs accept GGUF and Hugging Face SafeTensors model files directly as input alongside ONNX, so no conversion or export step is required. With Windows ML 2.9.42 Experimental, SafeTensors models run across text generation, speech recognition, and decision tasks, while GGUF files are supported for text generation, including via a llama.cpp loader with a CUDA GGML backend for GPU acceleration. Teams choosing between model formats for local AI apps can follow this kind of compatibility news on daily.dev."}},{"@type":"Question","name":"Which NVIDIA driver is needed for Windows ML DX CGC GPU acceleration?","acceptedAnswer":{"@type":"Answer","text":"Vendor-specific GPU acceleration for DX CGC in this initial release requires the NVIDIA GeForce 621.24 driver, or the equivalent 621.24 driver for RTX Spark. Support for additional GPU vendors is planned for future releases; meanwhile CPU and NPU workloads run through the Runtime APIs via ONNX Runtime-based execution instead of DX CGC. Developers setting up local GPU inference pipelines can keep driver and compatibility updates on daily.dev."}}]}
```

