For years, Windows on ARM has operated as a walled garden for graphics hardware. Qualcomm’s silicon dominates the platform, and Nvidia has never shipped a public ARM64 driver for its desktop GeForce lineup. That barrier just cracked. A developer has successfully run an Nvidia RTX 4060 on Windows 11 ARM by extracting a native ARM64 driver from Nvidia’s RTX Spark software, proving that the architecture can communicate with modern discrete GPUs.
The experiment, first spotted by Windows Latest and detailed in a translated video by a Bilibili creator known as VoidTech, pushes the boundaries of what Windows on ARM can actually do. While the setup is far from a plug-and-play solution, it confirms that Nvidia’s unified driver architecture can theoretically support multiple generations of RTX cards on ARM processors.
The Hardware and the Challenge
The test rig is far from a standard gaming PC. It is built around a Huawei Qingyun W510 workstation powered by a 24-core Kunpeng 920 processor. This chip relies on the ARMv8.2 instruction set, the same foundational architecture that powers Qualcomm’s Snapdragon X Elite in modern laptops. The system is equipped with 32GB of DDR4-2933 memory and the Nvidia RTX 4060 with 8GB of VRAM.
Getting Windows 11 to boot on this hardware was the first major hurdle. Because Huawei remains under US trade restrictions, the Qingyun W510 was never designed or officially supported for Microsoft’s operating system. It ships primarily with Linux-based distributions like UnionTech’s UOS. More importantly, the system lacks the ACPI tables that Windows requires to communicate with the processor, manage power, and enumerate connected devices.
To bypass this, the developer applied an open-source patch from the dixyes/d920s10 project. This patch manually corrects the ACPI tables, allowing the Windows kernel to recognize the Kunpeng 920 and initialize the hardware stack. Without this intervention, the operating system simply cannot initialize the platform, regardless of driver availability.
The Driver Extraction
Once Windows booted, the next obstacle was the GPU itself. Nvidia’s standard GeForce drivers are compiled exclusively for x64 processors. While Windows 11 can translate many x64 applications to run on ARM through its Prism emulation layer, it cannot translate kernel-level drivers. This is precisely why desktop Nvidia GPUs have never worked on Windows on ARM.
The breakthrough came from Nvidia’s own RTX Spark project, which recently began preparing N1x Windows 11 GPU drivers. Nvidia already ships a native ARM64 driver for its RTX Spark systems, but that software is locked to specific hardware. The developer extracted the ARM64 driver package from the RTX Spark software and installed it on the desktop RTX 4060, which uses a PCIe 4.0 x8 interface.
Windows successfully detected the card, enabling hardware acceleration, DirectX 12, and Vulkan. In a notable detail, the DirectX diagnostic tool reports the card as WHQL-certified, suggesting the driver package retains its digital signatures and compatibility metadata. Nvidia’s own nvidia-smi utility confirms the exact driver build at 591.33, paired with CUDA 13.1.
Because the driver is built on Nvidia’s unified architecture, the developer notes it could theoretically support other RTX graphics cards, including the RTX 20, 30, 40, and 50 series. This implies that Nvidia’s driver model is already hardware-agnostic at the kernel level, even if the company has not officially validated it for third-party ARM systems.
Performance and Gaming Reality
Despite the successful installation, the RTX 4060 on Windows 11 ARM is barely functional for daily use or gaming. The Kunpeng 920 suffers from weak single-core performance compared to modern desktop chips, and Windows must translate most games from x64 to ARM64 in real time. The developer estimates this translation overhead consumes roughly a quarter of the CPU’s total performance.
As a result, the GPU spends a significant amount of time idle, waiting for the CPU to feed it instructions. The developer ran several benchmarks to quantify the experience:
- Genshin Impact ran at 1080p across Fontaine and Mondstadt scenes, averaging around 20 FPS.
- The Black Myth: Wukong benchmark averaged about 21 FPS, with peaks at 33 FPS and drops as low as 3 FPS.
- The Star Wars Reflections demo ran at roughly 23 to 25 FPS.
Native ARM64 versions of 3DMark tests, including Night Raid, Solar Bay, and Speed Way, ran without translation overhead. Night Raid scored 43,530, Solar Bay reached approximately 32,370, and Speed Way hit 2,252. Time Spy remained an exception, as it still relies on x86/x64 translation, returning a 6,369 graphics score.
Productivity and creative workloads showed more promise. The x64 version of Blender successfully detected the RTX 4060 through Nvidia’s CUDA renderer. Task Manager confirmed the GPU was handling the rendering workload at roughly 89% utilization, though Blender itself was still running through the x64 translation layer. This workaround is currently necessary because a native Windows on ARM CUDA toolkit does not yet exist outside of WSL.
Ray tracing demos also functioned. The first and third Nvidia RTX demos rendered ray tracing and DLSS correctly, with the Star Wars Reflections demo enabling full ray tracing and reflections. The second demo refused to activate ray tracing, which the developer attributed to a GPU detection error or driver mismatch from the extraction process.
Anti-cheat software remains a hard barrier. Arknights: Endfield silently closed moments after launch, likely due to its ACE anti-cheat module or conflicts with the x64 translation layer. Additionally, the Huawei board cannot output video directly to a monitor when the RTX 4060 is installed. The developer had to rely on Sunshine and Moonlight to stream the desktop and gameplay to another device, which introduces additional latency and compression artifacts.
What This Means for You
This experiment does not change the Windows on ARM landscape overnight, but it removes a fundamental technical doubt. Nvidia’s driver architecture has already been adapted for ARM64 at the kernel level for its RTX Spark project. The fact that it can be extracted and made to communicate with a standard PCIe desktop GPU proves the software stack is not strictly tied to Nvidia’s own ARM silicon.
For Qualcomm, this reinforces the importance of its ongoing negotiations with GPU vendors. Windows on ARM’s long-term viability depends on native graphics support that does not require kernel translation. If Nvidia eventually ships an official ARM64 GeForce driver, it would unlock a massive portion of the existing PC gaming library for ARM laptops and desktops.
Until then, the RTX Spark project remains Nvidia’s primary vehicle for ARM graphics validation. The extracted driver serves as a proof of concept rather than a usable product, highlighting both the potential and the current limitations of running modern Windows GPUs on non-Intel, non-AMD processors.
How to Try It Yourself
This setup is not intended for everyday users. It requires a Huawei Qingyun W510 or similar Kunpeng-based ARM workstation, a compatible RTX desktop GPU, and advanced knowledge of ACPI patching and driver extraction. The process involves modifying system firmware tables, extracting proprietary driver packages, and accepting significant performance penalties from x64 translation.
If you are running an ARM-based Windows PC today, your options remain limited to integrated Qualcomm Adreno graphics or officially supported peripherals. Enthusiasts interested in ARM64 graphics development should monitor Nvidia’s RTX Spark driver releases and Qualcomm’s developer documentation for official ARM64 GPU validation paths.
Source: Windows Latest
Over to you: Will Nvidia ever ship official ARM64 GPU drivers for Windows, or will enthusiasts keep extracting them from RTX Spark builds?
You may also like
Microsoft Pulls the Plug on Copilot Deep Research and Podcasts Next Month
Microsoft brings AMD’s Helios platform and EPYC processors to Azure for AI and HPC workloads
Linus Torvalds Defends AI in Linux Kernel as Microsoft’s AI-Code Strategy Pays Off
Gears of War: E-Day partners exclusively with NVIDIA for PC graphics
