Sign in
← AI PulseSep 9, 2026

Deep · news · Single-source brief

CUDA Toolkit 13.4 Adds Windows on Arm Support and Rubin Architecture Preview

NVIDIA's CUDA Toolkit 13.4 extends application development to Windows on Arm and provides early functional support for the NVIDIA Rubin GPU architecture.

By Illumora Editorial

Source · Sep 9, 2026, 8:24 PM · On Illumora · Sep 9, 2026, 8:28 PM

Media from the primary source — shown here so you can stay on Illumora.

Rewritten from one allowlisted primary — not independent enterprise reporting. Lanes →

Brief drafted by Illumora’s editorial model from the linked primary source. Ops desk reviews flagged pieces. How we write →

Read the source →NVIDIA Developer Blog — CUDA Toolkit 13.4 Adds Windows on Arm Support and Greater Control over Shared GPUs | NVIDIA Technical Blog
Save

NVIDIA has released CUDA Toolkit 13.4, introducing support for Windows on Arm and a preview of the NVIDIA Rubin architecture. This update expands CUDA application development beyond Linux on Arm platforms and allows developers to begin porting applications to the next-generation GPU architecture.

The toolkit also enhances GPU management capabilities with Multi-Process Service V3, expands CUDA Python and CCCL functionality, and updates NVIDIA Nsight developer tools and core math libraries.

Key Points

  • CUDA Toolkit 13.4 adds support for Windows on Arm, extending CUDA application development from Linux on Arm.
  • The release provides preview functional support for the NVIDIA Rubin architecture (compute capability 107), enabling early application porting.
  • Multi-Process Service V3 introduces a modernized control layer with a scriptable CLI, named server instances, TOML configuration, and cgroup-integrated GPU memory limits.
  • CUDA Compute Fabric Transport (CFT) offers a transport-centric API for communication-library developers to move data across NVLink fabric using named logical endpoints.
  • CUDA Python's cuda.core 1.1.0 adds texture and surface programming, NUMA-aware managed memory, and .pyi type stubs.
  • CCCL 3.4 delivers a faster warp-specialized cub::DeviceScan on Blackwell GPUs, reaching up to 92% memory-bandwidth utilization.
  • CUDA SDK installers no longer bundle the NVIDIA driver, requiring separate installation via package managers.

Context

According to the NVIDIA Developer Blog, CUDA Toolkit 13.4 introduces early developer support for the NVIDIA Rubin GPU architecture, which is described as the next-generation GPU architecture powering the era of agentic AI. The Multi-Process Service V3 update aims to simplify the automation and management of shared GPU resources by providing a scriptable CLI, named server instances, and namespaces. It also integrates TOML configuration, streaming multiprocessor (SM) partition controls, and cgroup-integrated GPU memory limits for precise GPU partitioning in containerized environments.

CUDA Compute Fabric Transport (CFT) is designed for advanced applications and communication libraries to move data across NVIDIA NVLink fabric at scale. This approach reduces virtual-address pressure in large multi-GPU systems and supports unicast and multicast communication patterns. CFT is available through the CUDA Driver API and is intended for communication-library developers requiring specific functionality not available from higher-level libraries like NVIDIA NCCL or NVSHMEM.

Why It Matters

This release offers developers new capabilities for targeting diverse hardware platforms and managing GPU resources, which can impact application performance and deployment strategies. The early access to the Rubin architecture allows for proactive adaptation to future hardware, while enhanced GPU management tools provide finer control over resource allocation, crucial for optimizing workloads in containerized and multi-tenant environments.

What To Do

  • Download CUDA Toolkit 13.4 to access Windows on Arm support and the Rubin preview.
  • Review the MPS V3 quickstart to begin using the modernized GPU resource management capabilities.
  • Explore the cuda.core 1.1.0 release notes for a complete list of Python API enhancements.
  • Consult the CUDA Programming Guide for details on CUDA Compute Fabric Transport and locality domains.

Keep Exploring

/atlas/image-models