Microsoft tests Windows AI features on RTX GPUs, not just NPUs It's an experimental release for now, but probably will hit mainstream in a few months. , PDT Nvidia RTX GPUs through experimental Windows App SDK updates.This shift enables millions of older PCs with powerful GPUs to access AI tools like text summarization and image upscaling previously exclusive to newer Copilot+ devices.The change represents Microsoft’s more inclusive AI strategy, allowing broader Windows 11 device compatibility for local AI processing tasks. An under-the-hood change in Windows seems to signal the further deterioration of Microsoft’s Copilot+ branding, which, at least historically, depended solely on NPUs as the engine of local PC AI. Now, PCs with dedicated GPUs will have access to those features. An experimental release of the Windows App SDK on Github now allows certain AI-specific features to run on Nvidia RTX GPUs, rather than solely depending on an integrated NPU. The Windows App SDK 2.2 governs features like text summarization, text rewriting and editing, code generation, and others. It might also allow the Microsoft Photos app’s upscaling or Super Resolution feature to run on PCs with a moderately powerful GPU, as well as enabling “erase” and object extraction using AI. For over two years, we’ve known that GPUs generate the most AI TOPS within the PCs, even if the NPU is the most efficient AI engine for features like Windows Studio Effects, Paint’s Cocreator, and more. But Microsoft’s Copilot+ PCs originally looked first for the NPU and virtually ignored the GPU when supporting those features. That’s changing, apparently. For now, Microsoft’s “support” for GPUs requires downloading the experimental version of the WinAppSDK 2.2 Experimental 9 and pairing it with an Nvidia GeForce RTX 30 series GPU with at least 6GB of VRAM. You’ll also need a PC running a Windows Insider build. You’ve got to enable Developer Mode, too. That’s dedication. Still, as Windows Latest points out, it’s less about what it enables and more about what it signals: Microsoft continues to chip away at what was a monolithic approach to AI–the NPU or bust. In an interview with Andrew Hill, Microsoft’s corporate vice president of Surface, at Microsoft’s Build conference last week, my colleague Alaina Yee noted how the company was becoming increasingly open to running local AI models where it made sense and relying on the cloud for other AI tasks when needed. Likewise, Microsoft’s experimental Windows App signals that it’s trying to bring its AI capabilities to a broader range of Windows 11 devices, a more common-sense approach. As much as some segments of the internet tend to frown on generative AI as a replacement for creation, the ability to use AI as a tool seems to be much more broadly accepted. Millions of older PCs lack an NPU but have access to a robust GPU. A move like this is a no-brainer, even if it may take a few months to become more broadly available to the Windows community at large. : Mark Hachman, , Mark has written for for the last decade, with 30 years of experience covering technology. He has authored over 3,500 articles for alone, covering PC microprocessors, peripherals, and Microsoft Windows, among other topics. Mark has written for publications including PC Magazine, Byte, eWEEK, Popular Science and Electronic Buyers' News, where he shared a Jesse H. Neal Award for breaking news. He recently handed over a collection of several dozen Thunderbolt docks and USB-C hubs because his office simply has no more room. Recent stories by Mark Hachman: Apple’s new Siri just works. Why can’t Copilot? Windows Insider meetups are back. I went to one and so should you Microsoft is building a ‘quiet’ Windows, starting with animated grass
