#NVIDIA
68 related articles
NVIDIA Nemotron 3.5 Transcribe Released: 40 Languages, 1-Hour Recordings, Open Weights
NVIDIA releases Nemotron 3.5 Transcribe speech model with 40 languages, up to 1-hour recordings, prompt-guided transcription, speaker labels and audio QA; weights open in BF16/FP8/NVFP4, with day-0 vLLM and NeMo Speech serving.
NVIDIA RTX Spark: Rebuilding Game Neural Rendering with Local AI Compute
NVIDIA launches RTX Spark, a Windows PC platform built on Blackwell RTX architecture for AI development, content creation, and gaming, featuring local neural rendering and AI characters in thin laptops and compact desktops.
NVIDIA Sentry Explained: How Runtime Security for AI Agents Works in Practice
NVIDIA Sentry pushes AI agent security controls down to infrastructure layers such as BlueField DPUs, managing traffic, monitoring behavior, protecting data, and enforcing firewall policies at runtime to close the action-level overreach risks that model-layer guardrails cannot cover.
NVIDIA DSX POD: Bringing AI Compute Online Faster
NVIDIA DSX POD targets the system integration challenge before AI factories go live, aiming for faster, more predictable production-grade AI capacity. This article examines its positioning, impact on operations teams, and open questions.
Blackwell RTX and Grace CPU Laptops: OEM Engineering Trade-offs
NVIDIA's Blackwell RTX + Grace CPU RTX Spark laptops claim 128GB unified memory and 1 petaflop, but what engineering trade-offs do OEMs face in thermals, supply chain, and power?
Jensen Huang's Taipei NVIDIA Party: Lex Fridman Witnesses a Night of Gratitude in the Chip Empire
Lex Fridman documents NVIDIA's Taipei employee party: Jensen Huang personally serves food to media as thousands line up. He rarely shares observations on Taiwan's supply chain culture and college memories of a five-cent weekly allowance.
NVIDIA Releases Nemotron 3.5 Super VL: A 120B Open-Source Multimodal Agent Model
NVIDIA releases Nemotron 3.5 Super VL, a 120B-parameter open-source multimodal agent model with 12B active params, 1M context, and NVFP4 weights runnable on a single DGX Spark.
Meta Open-Sources Muse Spark: 816B Coding Model Lands on NVIDIA GB300
Meta open-sources Muse Spark Open: an 816B MoE coding model with 42B activated parameters, adapted for NVIDIA GB300 NVL72 and built for enterprise Agentic Coding.
NVIDIA Dynamo KV Routing in Practice: Smart Routing Configuration Cuts TTFT by 40%
A detailed guide to configuring NVIDIA Dynamo's KV-aware smart routing, covering aggregated and disaggregated serving modes, core parameter tuning, KV event stream architecture, and multi-backend support to achieve 40% lower TTFT and 25-30% higher throughput.
NVIDIA NeMo Gym Optimizes Chip Design AI Agent Reliability
NVIDIA NeMo Gym improves chip design AI Agent reliability through targeted training, solving hallucination output issues while optimizing tool-calling efficiency to move Agents from demo to production-ready.
NVIDIA Cosmos Lab on Open Models and the Future of Physical AI
NVIDIA Cosmos Lab VP Ming-Yu Liu discusses on the Practical AI podcast how open models lower barriers to physical AI, how world models give AI physical intuition, and the data bottlenecks facing physical AI.
CUTLASS Python: NVIDIA's New Tool for Accelerating GPU Kernel Development
NVIDIA engineer introduces CUTLASS for Python updates: new Qt extensions, CUTLASS primitives, compiler diagnostics, and static checks for high-velocity GPU kernel development for both developers and AI agents.
5 Steps to Build SimReady Assets for Robot Simulation: A Practical Guide with Frontier AI Models
NVIDIA's 5-step workflow for building SimReady robot simulation assets using Frontier AI Models — covering CAD-to-OpenUSD conversion, material setup, collision geometry, and physics annotation.
Building Reliable Data Analysis Agents: Lessons from NVIDIA KGMON at KDD Cup
NVIDIA KGMON placed 2nd at KDD Cup by simplifying the agent harness for reliability. Learn key engineering lessons for building production-ready data analysis agents.
DOCA GPUNetIO: Unifying GPU-Initiated Networking Across the NVIDIA Software Stack
NVIDIA DOCA GPUNetIO makes network communication a first-class GPU-controlled operation, unifying GPU-initiated networking across the stack to reduce CPU bottlenecks and latency for AI and HPC.
NVIDIA Green Contexts: Fine-Grained Control Over GPU Resource Allocation
NVIDIA Green Contexts lets developers programmatically partition GPU SM resources within a single process, isolating latency-sensitive and throughput-intensive workloads for predictable performance.
NVIDIA Open-Sources OpenShell: A Secure, Private Runtime for AI Agents
NVIDIA open-sources OpenShell, a Rust-based secure and private runtime for autonomous AI agents, gaining nearly 6,000 GitHub stars in one week. A deep dive into its design and significance.
Build a Visual AI Agent in 30 Minutes: NVIDIA VSS Blueprint in Action
Build a visual AI agent in 30 minutes for ~$3 using NVIDIA VSS Blueprint 3.3. Cosmos monitors cameras, Nemotron generates reports — fully local, just cents per task.
Micro Center Requires No-Export Declaration for RTX 5090 Purchases, Targeting GPU Smuggling to China
Micro Center reportedly requires customers to sign a No-Export Declaration when buying the RTX 5090, targeting GPU smuggling to China amid tightening AI compute export controls.
NVIDIA DOCA Agent Skills: Accelerating BlueField Application Development
NVIDIA DOCA Agent Skills injects BlueField DPU domain knowledge into AI agents, bridging the gap general-purpose models face in specialized infrastructure development.