DirectML: All about the Artificial Intelligence revolution in Windows and gaming

In recent years, machine learning and artificial intelligence have taken on an increasingly prominent role in the world of technology, especially in the realm of gaming, creative software, and the Windows operating system. However, while terms like DLSS, ray tracing, and hardware acceleration have been gaining traction for some time, there's a key component that connects this entire mechanism within the Microsoft ecosystem: DirectML . If you've ever wondered what DirectML is, what it's used for, and how it's becoming the secret engine behind many current innovations, here's the most comprehensive and up-to-date explanation.

DirectML isn't just a buzzword; it's an essential technology designed to unlock the full potential of modern GPUs and NPUs in artificial intelligence and machine learning tasks, whether in next-generation video games, multimedia editing apps , or advanced Windows features. Understanding DirectML today means understanding a significant part of the technological revolution taking place in PCs, consoles, and Windows devices.

What is DirectML and why is it important?

DirectML (Direct Machine Learning) is a low-level API created by Microsoft specifically to accelerate machine learning tasks on Windows systems using the DirectX 12 standard. It's a library that allows developers to leverage the power of graphics processing units (GPUs) and non-processing units (NPUs) to efficiently run artificial intelligence models. This API is intended not only for game development but also for all types of applications, engines, middleware, and environments that require advanced machine learning capabilities.

The raison d'être of DirectML lies in the integration and acceleration of artificial intelligence within the Windows ecosystem . It allows complex algorithms to leverage the immense parallelism and computing power offered by modern GPUs and, more recently, the NPUs incorporated into many current systems, rather than relying solely on the central processing unit (CPU). This makes tasks such as image processing, graphics enhancement, predictive model acceleration, and advanced task automation significantly faster, more efficient, and more accessible to a wider range of users and developers.

How does DirectML work in the Windows environment?

DirectML acts as a bridge between machine learning frameworks (such as TensorFlow or PyTorch) and DirectX 12 compatible hardware . Its implementation is deeply integrated into Windows 10 since version 1903 (build 10.0.18362) and into Windows 11 , making it easy for any compatible application to take advantage of this hardware acceleration.

The great advantage of DirectML is that it offers a consistent and simple API for developers, who can create and deploy AI inference models across a wide variety of devices and architectures . Whether we're talking about GPUs from AMD, Intel , NVIDIA , Qualcomm, or dedicated NPUs, DirectML's goal is to unify calls and enable maximum available performance. By using DirectX 12 as its foundation, it allows you to fully utilize your graphics hardware, with very low latency and unbeatable efficiency for tasks such as image supersampling, pattern recognition, and automatic multimedia classification.

In practical terms, DirectML has become the standard for accelerating AI frameworks on Windows . Microsoft promotes it as an integration channel, especially for use with ONNX Runtime (a runtime environment for standard AI models), TensorFlow, and PyTorch, allowing applications and games to achieve vital improvements in speed and processing power without relying on specific hardware from a single brand.

Real-life use of DirectML: current applications and examples

Far from being a theoretical technology, DirectML is already used in numerous creative and professional solutions, most notably in the new Windows Copilot+ PCs . These devices, equipped with advanced architectures featuring the latest generation of NPUs and GPUs, take full advantage of DirectML's capabilities to execute intelligent tasks locally and with ultra-efficiency.

Among the applications that have already incorporated DirectML are some heavyweights in creative and productivity software:

• Adobe Premiere Pro: Uses DirectML for automatic audio clip tagging, instantly classifying each file based on its nature (music, dialogue, effects, ambience), which greatly speeds up editing.
• Capture One: has incorporated two AI-enhanced features with DirectML, such as Match Look and AI Crop, making it easier for photographers and creatives to work without losing precision.
• Affinity Photo 2: Leverages DirectML and the Qualcomm Hexagon NPU for automatic object and subject selection in photography, eliminating the need for manual masking and making editing much more efficient.

In the gaming sector, DirectML has been key to the research and evolution of techniques such as intelligent supersampling and improving graphics performance without sacrificing quality. For example, when Microsoft unveiled the Xbox Series X|S, DirectML was identified by developers as one of the console's greatest strengths, enabling radical performance improvements through techniques like "ML super resolution," art style transfer, and the development of emergent gameplay experiences.

DirectML's relationship with Ray Tracing, DLSS, and the competition

One of the most frequently asked questions is whether DirectML competes directly with technologies like NVIDIA's DLSS or if it's an alternative to ray tracing . The short answer is that DirectML and DLSS, while sharing certain goals (improving quality and performance in video games through AI), are not exactly the same.

DLSS (Deep Learning Super Sampling) is a proprietary NVIDIA technology that uses deep learning to generate higher-resolution images from fewer pixels rendered by the GPU, improving FPS and visual quality. Although only available for RTX cards, the concept behind DLSS is based on machine learning techniques that Microsoft is also seeking to democratize through DirectML.

DirectML, for its part, is an open API integrated into Windows for any DirectX 12 compatible hardware . It allows developers to access a range of AI acceleration capabilities, from intelligent image upscaling (Super Resolution) to game engine optimization, texture generation, speech recognition, and automatic photo enhancement. Therefore, AMD, Intel, and NVIDIA can all use DirectML, opening the door to alternatives to DLSS on platforms like AMD GPUs with Radeon ML and similar technologies.

Ray tracing and AI also go hand in hand thanks to DirectML . With the arrival of DirectX 12 Ultimate, technologies like DXR (DirectX Ray Tracing) and DirectML work together so developers can access both realistic light ray simulation and AI applied to graphics, generating more realistic, faster, and more efficient images.

For example, AMD has worked on solutions like DirectML Super Resolution , enabling intelligent upscaling (upgrading a 540p image to crisp 4K while reducing aliasing) on ​​its RX 6000 and Xbox cards, following an open and multi-platform philosophy that competes with DLSS but is not limited to a single manufacturer.

DirectML integration into hardware: NPUs, GPUs, and the next generation of PCs

The most significant recent development is the integration of DirectML not only into GPUs but also into NPUs (Neural Processing Units) . Until now, DirectML was primarily used to leverage the power of graphics cards from AMD, NVIDIA, and Intel. However, with the arrival of Intel Core Ultra processors (Meteor Lake) and their AI Boost NPU, along with the Samsung partnership, DirectML has expanded to take full advantage of chips dedicated exclusively to AI tasks within the PC.

This represents a revolution compared to the previous exclusive reliance on the GPU . Now, tasks such as language model inference, real-time analysis, and multimedia classification can be outsourced to these NPUs, freeing up even more resources and improving power efficiency. Support has initially arrived on Intel and Samsung laptops running Windows 11, requiring a combination of software components (DirectML 1.31.1, NuGet, ONNX Runtime 1.17, and updated Intel drivers ) to function correctly.

Although AMD's Ryzen AI support is currently limited, it's clear the industry is moving toward greater DirectML integration across all types of hardware. Microsoft's goal is to ensure no developer or user is left behind in the AI ​​revolution, and in the near future, the new Ryzen 8040 series is expected to include advanced DirectML support and its NPU acceleration capabilities.

GPU Performance and Optimization Comparison: When is DirectML Worth It?

One of the most discussed topics regarding DirectML is its actual performance compared to other native GPU solutions . According to tests performed on different systems, for example using TensorFlow and DirectML in the Windows Subsystem for Linux (WSL), execution time can vary significantly depending on the GPU and drivers used.

For example, on a system with a Ryzen 3600 processor and an AMD RX 6600 XT graphics card, the test took approximately 139 seconds, while on an Nvidia RTX 3060 with native drivers, the time was 97 seconds. This demonstrates that, although DirectML is a powerful and flexible tool, there may still be situations where the pure performance of a native driver (especially on NVIDIA cards) is superior. However, the difference tends to decrease with each update and expansion of DirectML support, particularly on AMD hardware or in scenarios where cross-platform compatibility and access to intelligent features from standard AI frameworks are prioritized.

It is not always a good idea to use DirectML, especially in these cases:

• If you have a very powerful GPU but no DirectML support, the native driver may offer better performance.
• For extremely complex and specific machine learning tasks, you may need more specialized solutions or libraries optimized directly for your hardware.
• If your development doesn't require AI or ML, DirectML offers no added value, and you can opt for other tools tailored to your workflow.

Current advantages and limitations of DirectML

Among the main benefits of DirectML are its seamless integration with Windows, compatibility with a wide range of hardware, and its ability to make AI accessible and efficient for any user or developer . It allows for the acceleration of both productivity applications and all types of video games, without requiring extensive hardware knowledge, provided the system meets the minimum DirectX 12 requirements and has adequate support.

Furthermore, DirectML is the channel through which developers can anticipate the future of on-premises AI, avoiding exclusive reliance on the cloud or proprietary solutions . Thanks to its integration with technologies such as ONNX Runtime, TensorFlow, PyTorch, and other popular AI libraries, the range of possibilities is as vast as the imagination of engineers and creatives.

However, there are still some disadvantages or barriers to entry :

• Performance in some cases still lags behind native drivers optimized for a specific GPU brand, although the gap is narrowing every year.
• NPU integration is still in its early stages, meaning only certain hardware models, versions, and drivers can take full advantage of these capabilities.
• There may be compatibility limitations for highly complex or specialized deep learning models, in which case it may be necessary to use alternative libraries.

Requirements, installation and startup of DirectML

To take advantage of DirectML on Windows, the first requirement is a DirectX 12 compatible system and the appropriate operating system version (Windows 10 build 1903 or later, or Windows 11) . From there, installation and usage are typically tied to the software you intend to use (for example, TensorFlow or PyTorch) and the corresponding GPU or NPU driver.

In advanced scenarios, such as using Windows Subsystem for Linux (WSL) and ML acceleration in Linux environments under Windows, the steps are typically as follows:

1. Enable the WSL subsystem and Virtual Machine Platform from Windows Features.
2. Install a Linux distribution (such as Ubuntu) from the Microsoft Store.
3. Set up a development environment like MiniConda for managing Python environments.
4. Install the appropriate version of TensorFlow or PyTorch and integrate DirectML support according to the official documentation.

For NPU acceleration, it's essential to have the latest version of DirectML, ONNX Runtime, and the most up-to-date drivers for the corresponding NPU (for example, Intel AI Boost in Meteor Lake) . Not all models or tasks are supported in the initial releases, so it's crucial to consult the release notes and supported model lists from each hardware vendor.

The Future of DirectML: AI at the Heart of the Windows Ecosystem

Microsoft continues to invest heavily in DirectML as a driver of innovation in artificial intelligence for Windows . The development of Windows Copilot+, its integration with new chips from Intel, Qualcomm, and AMD, and the opening of the platform to all types of developers foreshadow unstoppable expansion for DirectML in the coming years.

The DirectML team is working to add more features and make it easier for any developer to leverage AI in their applications , accelerating future innovations and democratizing the power of artificial intelligence across desktop, web, and mobile devices . The trend is clear: more performance, fewer technical barriers, and an increasingly seamless and personalized experience for Windows users.

Related articles:

Gaming and AI: What the PC revolution will look like in 2025

Isaac

Passionate writer about the world of bytes and technology in general. I love sharing my knowledge through writing, and that's what I'll do on this blog, show you all the most interesting things about gadgets, software, hardware, tech trends, and more. My goal is to help you navigate the digital world in a simple and entertaining way.