Nvidia Intelligence Hub
Real-time accumulated intelligence, primary research, and executive digests for Nvidia.
Nvidia open sources cuFile API
Nvidia has open-sourced its cuFile API, a key part of its GPUDirect Storage stack, to accelerate direct data movement between high-speed storage such as NVMe drives and GPU memory with millisecond-level latency. Originally launched in 2021 alongside CUDA Toolkit 11.4, cuFile bypasses CPUs and main memory using direct memory access, reducing bottlenecks and GPU starvation for large-scale AI workloads, including retrieval-augmented generation and agentic AI systems. Nvidia also unveiled the Storage-Next initiative with 40 storage and flash vendors and introduced its SCADA architecture, which seeks to balance ultra-fast GPU data access with Linux-based protections against clobber and potential security gaps.
Read Original Source ↗NVIDIA releases Molt PyTorch agentic RL framework
NVIDIA’s NeMo team has released Molt, an Apache 2.0 licensed, PyTorch-native agentic reinforcement learning framework designed as a compact, roughly 8.6K-line codebase for researchers and AI coding assistants. Molt simplifies rapid algorithm iteration by keeping agents as plain Python programs, integrating Ray for placement, vLLM for rollouts, and NVIDIA AutoModel with FSDP2 for training, while maintaining token-exact trajectories and policy-version consistency. Although framed as research infrastructure rather than a production service, Molt targets labs with multi-node H100 or H200 access, supporting applications like tool-use agents, vision-language environments, LLM-as-judge reward loops, and large-scale on-policy distillation.
Read Original Source ↗