NVIDIA CUDA 13.4 Adds Windows-on-Arm Support for RTX Spark

NVIDIA has released its CUDA 13.4 Toolkit. The update provides key support for Windows-on-Arm systems ahead of the launch of RTX Spark laptops next month.

This release allows developers to begin porting and validating applications for the Arm64 platform on existing hardware. It marks a major milestone for the upcoming RTX Spark roadmap.

Key Updates for Developers

The CUDA 13.4 toolkit focuses on broadening platform support and providing new tools. The main highlights include enabling CUDA to run on existing Windows on Arm machines. This is targeted at helping developers prepare Windows Arm64 applications for the October arrival of RTX Spark laptops.

Another significant feature is preview support for Vera Rubin CPUs and GPUs. This early access allows developers to work with the CUDA platform capabilities for the next-generation architecture. The report states this is particularly important for maintainers of key libraries.

Technical Enhancements

The update includes broader changes across the CUDA ecosystem. These span the compiler, PTX ISA, programming models, runtime APIs, libraries, and developer tools.

Specific new features are detailed. Multi-Process Service V3 introduces a modernized control layer with a scriptable command-line interface. It also offers named server instances and TOML configuration. This update includes cgroup-integrated GPU memory limits for precise partitioning in containerized environments.

CUDA Compute Fabric Transport provides a new API for communication-library developers. It is designed to move data across NVLink fabric using named logical endpoints and asynchronous operations.

CUDA Python also sees expansion. The cuda.core 1.1.0 release adds texture and surface programming support. It includes NUMA-aware managed memory and.pyi type stubs. Meanwhile, cuda.compute 1.1 enables ahead-of-time compilation of CCCL algorithms for multiple GPU architectures.

Focus on Future Hardware

Core math libraries in the toolkit now have functional support for the Rubin GPU architecture. They also include Windows on Arm support tailored for the N1X Laptop ecosystem.

The Vera Rubin preview is a forward-looking component. CUDA Toolkit 13.4 adds functional support for the NVIDIA Rubin architecture, identified as compute capability 107. This preview enables developers to begin porting applications before general availability in a future toolkit release. NVIDIA describes Rubin as the next-generation GPU architecture powering the era of agentic AI.

The NVIDIA CUDA 13.4 toolkit is now available for download.