NVIDIA’s October 27, 2026 developer update describes new RTX Spark support for Unreal Engine 5.8, Nsight tools, game AI and more. For developers working on Windows on Arm, the practical CUDA detail is narrower: NVIDIA lists support for specific components in CUDA Toolkit 13.4 Update 2 and requires a separate RTX Spark driver, version 616.41 or later.
For background on the platform and DLSS 5, see NeoTeo’s earlier RTX Spark coverage. The newer announcement focuses on developer tools. NVIDIA describes RTX Spark as a Windows platform with a Blackwell RTX GPU, Grace CPU and up to 128 GB of unified memory, rather than a single laptop configuration.
Which RTX Spark tools support Windows on Arm?
NVIDIA lists different Windows-on-Arm support across the DLSS Unreal Engine plugin and three Nsight tools. The versions and scopes are not interchangeable: some cover native Arm64 applications, while others also account for emulated x64 or Arm64EC targets.
| Tool | Version | Stated Windows-on-Arm scope |
| DLSS Unreal Engine plugin | 8.9.0 | Unreal Engine 5.8 support, native ARM64 binaries, and UE 5.8 build and packaging support |
| Nsight Graphics | 2026.3 | Windows-on-Arm installer; supports native Arm64, emulated x64 and Arm64EC target applications |
| Nsight Systems | 2026.5 | One Windows installer for native Arm64 and emulated applications |
| Nsight Aftermath SDK | 2026.3 | Windows-on-Arm SDK for native Arm64 applications; the x64 SDK handles crash dumps from emulated x64 applications |
| CUDA Toolkit | 13.4 Update 2 | Windows Arm64 support for selected components, including NVCC, CUDA Runtime, CUPTI and Profiler API |
Arm64 refers to software built for Arm processors. Emulated x64 applications run through Windows’ compatibility layer, while Arm64EC lets native Arm code work alongside emulated x64 code. NVIDIA’s listed support gives developers specific targets to check when planning a build; it does not establish compatibility for every plugin or dependency in a project.
NVIDIA also lists eight RTX Kit technologies for RTX Spark: Neural Texture Compression, Texture Filtering, Neural Shaders, Mega Geometry, Real-Time Denoisers, Character Rendering, Global Illumination and Opacity Micro-Maps. These are graphics and rendering technologies, separate from the Windows-on-Arm support scopes in the table.
CUDA Toolkit and the separate driver
Installing CUDA Toolkit does not install the RTX Spark Windows driver. NVIDIA’s CUDA 13.4 Update 2 release notes say the driver is no longer bundled with the toolkit and specify driver version 616.41 or later for RTX Spark on Windows. The notes identify R615 as CUDA 13.4’s driver branch. Developers need to account for the driver as a separate installation requirement.
The release notes’ component matrix lists Windows Arm64 support for NVCC, the CUDA Runtime, CUPTI and the Profiler API, among other components. It lists CUDA GDB for Linux architectures, not Windows Arm64. This component-level support does not establish native Windows-on-Arm compatibility for PyTorch, Triton, TensorRT or every third-party package. NVIDIA’s CUDA Toolkit 13.4 Update 2 release notes provide the component and driver details.
DLSS 5, Nemotron and MotionBricks
NVIDIA says its October DLSS 5 model is 2× faster than the September version, with half the GPU overhead and 20% less VRAM use at 4K while maintaining comparable visual fidelity. Those figures are NVIDIA’s comparison of the two models.
For game characters, NVIDIA describes Nemotron 3.5 Bolt E4B 4GB as available through its In-Game Inferencing SDK for local experiences. The company lists multi-turn tool use, configurable reasoning, a 64k context window and support for 35 languages; it says the base model was pretrained on more than 140 languages.
The MotionBricks update includes SOMA human-motion models, downloadable checkpoints, ONNX exports and a Newton-based reference implementation. NVIDIA says the Unreal Engine integration shown in its research demonstration is not part of the release.
