AI infrastructure and developer tool updates #133
Today's Letter
NVIDIA, CUDA Toolkit 13.4 released
- CUDA Toolkit 13.4 adds Windows on Arm support, extending CUDA development beyond Linux on Arm
- Preview support for NVIDIA Rubin GPUs introduces the compute capability 107 target for early porting
- Multi-Process Service V3 adds a scriptable CLI, named server instances, TOML configuration, SM partitioning, and cgroup-integrated GPU memory limits
- CUDA Compute Fabric Transport enables asynchronous data transfers across NVIDIA NVLink using named logical endpoints
- CUDA Python adds texture and surface programming, NUMA-aware managed memory, and type stubs through cuda.core 1.1.0
- SDK installers no longer bundle the NVIDIA driver, which must be installed separately
Source: developer.nvidia.com
Qwen3.8-Omni-Flash Launches
- Qwen launched Qwen3.8-Omni-Flash as a native omnimodal model for agentic audio and video workflows
- The model accepts text, image, audio, and video inputs with a 1M-token context window
- Target workflows include video editing, music video creation, film commentary, audiovisual summarization, and real-time conversation
- Qwen-MM-Plugins adds perception, tool use, and workflow execution for long-form audio and video
- Qwen-Live Harness was open-sourced as a runtime for continuous, real-time omnimodal interaction
Source: qwen.ai
More: byteshape.com · techgenyz.com · gigazine.net
Jocoletter curates AI, software, and product trends for developers and builders.
#NVIDIA #Qwen