NVIDIA's RTX Spark launches this fall, and the company has released the first preview of its native Windows on Arm drivers as a preview release.
NVIDIA RTX Spark Systems Shipping To Devs As First Preview Drivers Roll Out
NVIDIA's RTX Spark superchip for AI PCs is rolling out in a few months across various laptops and Mini PCs. These systems will offer the power of the Grace CPU architecture, along with a Blackwell GPU, combined with 128 GB of memory, all interconnected using high-speed fabric, on a single chip that is designed for AI.
Today, NVIDIA has started to roll out its first developer preview drivers for the RTX Spark platform, which starts with the 616.00 release. This specific driver aims at the Microsoft Surface RTX Dev box, which is a developer-first solution and offers a fully fanless design.
The CUDA 13.4 Toolkit further references the RTX Spark support as a "Windows on Arm" preview. It is stated that the preview includes compiler and build-system enablement for Windows Arm64 targets:
- Native Windows Arm64 development for the upcoming RTX Spark device.
- Cross-compiling Windows Arm64 CUDA applications using the Windows x86_64 Toolkit.
Starting with CUDA 13.4, the driver will not be bundled within the toolkit but rather released separately with NVIDIA's Developer Driver (R616 & above).
According to NVIDIA, developers can start porting their applications to Windows on Arm on an existing development system. Devs can also identify incompatible dependencies and bring up an initial Arm64 build so that their apps are ready to deploy on the multitude of RTX Spark PCs that are launching this fall.
NVIDIA also provides a step-by-step guide, which you can see here.
Recommended Steps
- Review your application and third-party dependencies for Arm64 support.
- Choose an Arm64 or Arm64EC porting strategy appropriate for your application.
- Build and test your application on Windows on Arm.
- Validate the NVIDIA and CUDA software paths used by your application.
- Test installation, updates, functionality, and performance.
- Validate on RTX Spark when supported hardware and software become available.
Known Issues
- Reduced CUDA transfer performance with pageable memory
- CUDA host-to-device transfers using pageable or unpinned host memory may deliver lower-than-expected performance.
- Workaround: Where possible, use page-locked host memory allocated or registered with cudaMallocHost, cudaHostAlloc, or cudaHostRegister.
- Temporary blank display during driver installation
- The built-in display may remain blank for approximately two minutes during an Express or Custom driver installation. The display should recover automatically, and installation will complete successfully. Do not restart or power off the system while the display is blank.
- Possible system instability during PyTorch build workflows
- Running PyTorch CI/CD workflows may trigger a GPU timeout, causing the system to become unresponsive or restart unexpectedly. This issue is under investigation.
- Nsight Copilot is not available in the Developer review
- Nsight Copilot is disabled and unavailable on Windows ARM64 in this developer preview.
In our hands-on with the first RTX Spark systems at Computex 2026, we got to see some impressive systems with even more performance within AI applications, content creation tools, and games. The RTX Spark will be NVIDIA's first full-fledged entry on the laptop and AI PC market, and so far, RTX Spark looks very capable, but there are definitely a lot of metrics that need to be discussed.
Also, we've seen a range of benchmarks appear in leaks, but those are based on very early engineering samples that aren't reflective of retail systems, so take them with a grain of salt, and hold on till Fall 2026 to see how the RTX Spark performs in real-world use cases.
About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.
Follow Wccftech on Google to get more of our news coverage in your feeds.
