started · updated
NVIDIA releases CUDA Toolkit 13.4 with Windows on Arm support
NVIDIA has released CUDA Toolkit 13.4, which introduces native development support for Windows on Arm64. While CUDA applications have previously been supported on Arm platforms via Linux, this update extends that capability to the Windows platform. Developers can now compile applications directly on Windows Arm64 systems or use the Windows x86-64 toolkit for cross-compilation.
The release also provides early preview support for NVIDIA’s upcoming Rubin GPU architecture, featuring compute capability 10.7. This allows software developers and library maintainers to begin porting and testing applications ahead of the hardware's general availability.
Additional technical enhancements include Multi-Process Service (MPS) V3, which offers improved management of shared GPU resources through a scriptable command-line interface, TOML configuration, and cgroup-integrated memory limits. The update also expands CUDA Python functionality and introduces CUDA Compute Fabric Transport for data transfers across NVLink-connected systems.
Entities
CUDA · Rubin · Windows on Arm