RTX Spark receives first native CUDA Toolkit for Windows Arm64
NVIDIA has released the CUDA Toolkit 13.4 Developer Preview with its first official CUDA development package for Windows on Arm64. The release targets upcoming systems based on the RTX Spark platform.
This is not the first version of CUDA for the Arm architecture. NVIDIA has supported Arm64 devices through Jetson and Linux Arm64 SBSA packages for years. However, previous releases did not provide a native Windows Arm64 CUDA Toolkit.
CUDA development arrives on Windows Arm64
CUDA Toolkit 13.4 includes compiler and build-system support for Windows Arm64 targets. Developers can compile CUDA applications directly on a Windows Arm64 system or cross-compile them using the Windows x86-64 version of the toolkit.
© NVIDIA NVIDIA is providing a local Windows Arm64 installer and Windows Arm64 binary archives. The preview contains CUDA headers, import libraries, runtime components and libraries required for compilation and linking. Drivers, platform firmware and operating system images are distributed separately.
RTX Spark hardware is not required for porting
NVIDIA says developers can begin porting applications using an existing Windows on Arm system. The company recommends checking third-party dependencies for Arm64 support and choosing between native Arm64 and Arm64EC builds before validating the software on RTX Spark hardware.
© NVIDIA The developer preview has several known issues. CUDA transfers using pageable memory may perform below expectations, while some PyTorch build workflows may cause GPU timeouts or system restarts. The display may also remain blank for around two minutes during driver installation, and Nsight Copilot is not available on Windows Arm64 in this release.
CUDA Toolkit 13.4 remains pre-release software and is not intended for production deployment or benchmarking. NVIDIA has not confirmed when Windows Arm64 support will move into a general-availability CUDA release. In fact, RTX Spark doesn’t even have an official release date yet.
