NVIDIA announced the release of CUDA 13.5, which includes general availability support for the NVIDIA Vera Rubin GPU architecture and extends Windows on Arm support to NVIDIA DGX Station. The release also introduces enhancements in locality domain partitioning and multi-process service capabilities.
What happened
NVIDIA has released CUDA Toolkit 13.5, which brings general availability support for the NVIDIA Vera Rubin GPU architecture, moving from early developer access to enterprise production readiness. Additionally, Windows on Arm support, which was introduced in CUDA 13.4, now extends to NVIDIA DGX Station for Windows. CUDA 13.5 also includes enhancements in locality domain partitioning, allowing applications to choose between uniform streaming multiprocessor (SM) partitions or partitions that include all available SMs. CUDA Tile gains multi-GPU communication capabilities and a public preview of SIMT interoperability, enabling Tile kernels to call device functions from libraries such as NVSHMEM, CCCL Coop, Warp, and MathDx.
Source: NVIDIA
