DigestAI news desk

Cut through the AI noise.

Hardware & Compute1 min read

NVIDIA releases CUDA 13.5 with Vera Rubin support and GPU improvements

The update brings general availability support for NVIDIA Vera Rubin architecture, moving beyond early developer access. It also expands Windows on Arm support to DGX Station for Windows and introduces new driver modes for Rubin GPUs, replacing the older TCC mode.

1 source

Key points

  • CUDA 13.5 adds general availability for NVIDIA Vera Rubin architecture and Windows on Arm support for DGX Station
  • New driver modes (WDDM, MCDM) replace TCC mode for Rubin GPUs, improving GPU scheduling and resource management
  • CUDA Tile enables direct GPU communication via NVLink/NVSwitch, and CUDA Python 1.2 introduces batched memory transfers

Key improvements include updates to Multi-Process Service (MPS) v3, enabling weighted timeslicing and better GPU resource sharing. CUDA Tile now supports direct GPU-to-GPU communication via NVLink and NVSwitch, while CUDA Python 1.2 adds batched memory transfers and flexible CUDA Graph updates. The release also includes new AI inference algorithms in CCCL 3.5 and a public preview of Nsight Perf SDK for AI. Users must install the R620 driver branch separately, as drivers are no longer bundled with the toolkit.

Full story from videocardz.com · by WhyCry · via Search: NVIDIAOpen source ↗

NVIDIA CUDA 13.5 released with Vera Rubin support and improved GPU resource management

videocardz.com · 4 November 2026

Loading the full article…

This text was published by videocardz.com and written by WhyCry. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Topics · follow one to build your own front page

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Hardware & Compute

All →

Related stories