Skip to main content
AmigoFish

NVIDIA Developer

NVIDIA DeveloperVideo

Format:
YouTube channel
New videos:
Several times a week
Typical length:
about 50 minutes, not counting Shorts
Latest video:
Not rated yet

About this show

If you’re building with NVIDIA technology, you’re in the right place.

Find practical tutorials, technical deep dives, demos and talks from NVIDIA engineers and the developer community. We cover CUDA, AI agents, generative AI, robotics, data science, computer vision, simulation, high performance computing and more.

Whether you’re just getting started or optimizing something already in production, subscribe for new videos to help you build.

Covers: AI development, nvidia technology, physical AI, robotics, generative AI, AI agents, computer vision, edge AI, large language models, machine learning tutorials

Videos come from the channel's public YouTube feed. The artwork is the newest video thumbnail. Last updated . Names, artwork and text belong to their publishers.

Recently on this show

Summary of recent videos, drawn from the publisher's feed.

NVIDIA Developer is a video channel for people building with NVIDIA technology, covering topics such as CUDA, AI agents, robotics, and computer vision. Recent videos include a Jetson livestream on building physical AI applications, a Mercedes-Benz factory talk on AI agents, and a Nemotron 3.5 Transcribe livestream.

Report a problem with this summary

Cosmos

Similar in subject

See the full Cosmos

Similar shows

See all Technology shows

Videos

  • Build Physical AI Applications with Open Models on NVIDIA Jetson | Jetson Holiday Stream

    Learn how to build Physical AI applications with open models on NVIDIA Jetson. In this livestream, community developers demonstrate their projects and walk through model selection, application architecture, and deployment on Jetson. Hear how they tackled implementation challenges, explore lessons you can apply to your own builds, and bring your questions for the featured developers.

    Not rated yet
  • GTC DC Dev Community Spotlight

    What does it take to put AI agents to work across industries? Join us from GTC DC to explore how developers approach specialized agents, connect models to enterprise knowledge, and address safety and deployment challenges. Through conversations and demos, we’ll connect the event’s themes in healthcare, financial services, and government to practical developer takeaways. Discover tools and resources for your next project, and bring your questions to live audience Q&A.

    Not rated yet
  • GTC DC Dev Community After Hours

    Explore the technology themes shaping GTC DC from open models and AI agents to AI safety and accelerated computing. Join us for a developer-focused conversation about their potential across healthcare, financial services, Physical AI, and government. Discover how these technologies connect to practical workflows, what to explore at GTC DC, and where to find opportunities to learn and build. Bring your questions and join the conversation live.

    Not rated yet
  • Mercedes-Benz × NVIDIA Cosmos — Tuscaloosa Factory

    · 1 min

    From detecting assembly defects to flagging missed steps in real time, @MercedesBenz is using AI agents powered by @NVIDIA Cosmos to improve production quality. Michael Schiebe, Board Member at Mercedes-Benz, shares how physical AI is helping build self-learning factories where machines adapt and work alongside people.

    Not rated yet
  • Helping Delivery Robots Prepare for the Unexpected

    · 1 min

    What if one delivery run could become a whole set of edge-case tests? Coco Robotics’ robots make deliveries through platforms like Uber Eats, DoorDash, Deliveroo, and Wolt. Using NVIDIA Cosmos, Coco turns real fleet video into new variations of a scene, so its robots can practice handling rare incidents before encountering them on the street. #Robotics #robotdelivery #NVIDIACosmos

    Not rated yet
  • Beyond Transcription: Nemotron 3.5 Transcribe | Nemotron Labs

    · 52 min

    Nemotron 3.5 Transcribe is here: an open-weight, promptable 30B mixture-of-experts model (about 4B active parameters) that brings transcription in 40 languages, speaker attribution, translation, and audio understanding into a single checkpoint. In this livestream, NVIDIA engineers and our partners at Hugging Face walk through what the model does, how to prompt it with context for domain-specific vocabulary, and how to adapt it with the Granary v2 dataset. Expect a live demo on recorded audio. What you’ll learn: - How to run Nemotron 3.5 Transcribe from Hugging Face with NVIDIA NeMo or transformers - How contextual prompting improves accuracy on names, jargon, and domain terms - How speaker attribution and translation work in one checkpoint - How to adapt the model to your own domain with Granary v2 - When to use an offline model for recorded audio versus a streaming recognizer for live use - Building speech pipelines or fine-tuning ASR? Drop your questions live, and our guests will answer them in real time.

    Not rated yet
  • Open-Model Indic Language AI with NVIDIA Nemotron 3.5 Super VL

    India is home to 1.4 billion people speaking 22 scheduled languages , yet most AI models are built for English. In this inaugural Nemotron Livestream India session, NVIDIA and IIT Madras AI4Bharat Lab present its partnership through advance Indic language AI using NVIDIA Nemotron 3.5 Super VL open models. The session covers the open Indic dataset (3 trillion tokens across 22 languages), a data-generation pipeline for turning curated English content into quality Indic training data, and an Indic LLM benchmark suite spanning 15+ evaluations across math, STEM, code and Indian-language-specific capabilities. Attendees will learn about Nemotron 3.5 Super VL, a multimodal open model built to process and reason across text and images, purpose-built for autonomous agents running high-volume, specialized tasks, is available now. In this Ask the Experts session , you’ll get direct access to the NVIDIA AI team who built it: the architecture choices, training techniques, and design principles behind a model optimized with open weights and recipes you can customize for your domain. What you will learn: - Understand the partnership to advance Indic language AI across 22 Indian languages using NVIDIA Nemotron 3.5 Super VL open models. - Learn how the open Indic dataset (3 trillion tokens), the data-generation pipeline and the Indic LLM benchmark suite (15+ evaluations) are being built and released openly for developers to use and extend. - How NVIDIA AI researchers trained Nemotron 3.5 Super VL for agentic and vision-language workloads - What the open weights, training recipes, and datasets make possible for domain- and task-specific customization - How to integrate Nemotron 3.5 Super VL into multi-model agentic pipelines

    Not rated yet
  • From AI Factories to Intelligent Cities

    · 18 sec

    Thank you for a great 3 days, Barcelona! 💚 At @smartcityexpo World Congress 2026, @NVIDIA and our partners showcased how cities can put AI to work in public services, transportation, and infrastructure.

    Not rated yet
  • Serving GLM-5.3 Across Two NVIDIA DGX Stations with vLLM

    · 6 min

    When a model is too large for one GPU, you can still run it locally. In this tutorial, we serve GLM-5.3, a 753-billion-parameter mixture-of-experts model, across two NVIDIA DGX Stations with vLLM and tensor parallelism, keeping every weight in GPU memory with no CPU offload. The GLM-5.3 NVFP4 checkpoint is about 464 GB, more than the 252 GB of memory on a single NVIDIA GB300 GPU. By connecting two DGX Stations with ConnectX-8 and splitting the model across both GPUs, the two systems serve one model through a single OpenAI-compatible API. You'll see how to prepare both DGX Stations and collect the network values for the RDMA connection, download the pinned checkpoint and copy it to the second Station, and launch the same vLLM script on both machines. We then verify the service: a health check on the API, an RDMA check that confirms the GPUs communicate directly over the network instead of falling back to TCP sockets, and nvidia-smi showing one vLLM worker on each GPU. Next, we send a chat request through the OpenAI-compatible endpoint and check the answer. Finally, we use NVIDIA AIPerf with SPEED-Bench Coding prompts to measure how per-user throughput and request latency change as concurrency increases. The video runs a quick test at concurrency 1 and 8, then shows the reference results at 1, 8, 16, and 32 concurrent requests. The companion post on the NVIDIA Technical Blog covers the full benchmark and every command used in this walkthrough. Running a frontier-scale model on your own hardware is useful for working with sensitive data, prototyping agents locally, and iterating on prompts without per-token costs. To get started, connect your two DGX Stations with the Connect Two DGX Stations playbook, then follow the step-by-step commands in the companion blog. ➡️ Learn more: Connect Two DGX Stations playbook: https://build.nvidia.com/station/connect-two-stations 📝 Tech blog: https://developer.nvidia.com/blog/when-one-gpu-isnt-enough-serving-glm-5-3-across-two-nvidia-dgx-stations-with-vllm 📥 Download: GLM-5.3 NVFP4 on Hugging Face: https://huggingface.co/nvidia/GLM-5.3-NVFP4 Chapters: 00:00 Introduction 00:16 Why one GPU isn't enough 00:55 Step 1: Prepare the two DGX Stations 01:40 Step 2: Download and copy the model 02:21 Step 3: Launch the distributed vLLM service 03:45 Step 4: Verify serving and send a request 05:01 Step 5: Benchmark concurrency with NVIDIA AIPerf 06:20 Recap

    Not rated yet
  • Build Training-Ready Datasets in Two Prompts Featuring Encord and Voxel51 | Cosmos Labs

    Build diverse, training-ready datasets for physical AI with just two prompts using NVIDIA Cosmos Data Factory skills and blueprints. See how one prompt sets up your environment and another runs your chosen data workflow. Join NVIDIA, Voxel51, and Encord to explore how data curation, augmentation, generation, and labeling can expand training coverage for autonomous driving, robotics and vision AI agents. Voxel51 will share how Porsche curates and augments autonomous vehicle data powered by NVIDIA Cosmos Data Factory. Encord will walk you through its work with ANYbotics on synthetic data for anomaly detection in robotic inspection also powered by Cosmos Data Factory. What You'll Learn: - Set up and run a Cosmos Data Factory blueprint - Augment existing data and generate scarce training examples. - Explore customer use cases in autonomous driving and robotic inspection. Have questions about how to post-train and deploy NVIDIA Cosmos 3? Drop them live — the NVIDIA team will answer them in real time. Access more NVIDIA Cosmos developer resources and join our developer community: 📄 Read the Technical Blog → https://nvda.ws/4c6kK3R ⬇️ Download Cosmos on Hugging Face → https://huggingface.co/collections/nvidia/cosmos3 📚 Explore Models & Datasets on GitHub → https://github.com/nvidia/Cosmos 👥 Join the Cosmos Community → https://discord.com/invite/nvidiaomniverse

    Not rated yet
  • Accelerate Visual AI Agents on NVIDIA Jetson with VSS 3.3 and Cosmos

    Building visual AI agents at the edge means bringing together video understanding, application services, and device-specific configuration. In this hands-on livestream, we’ll bring together NVIDIA Blueprint for video search and summarization (VSS) 3.3, Cosmos, and Jetson agent skills in a practical application development workflow on the NVIDIA Jetson edge AI platform. We’ll start with Jetson Agent Skills to help a coding agent inspect the target hardware, check JetPack and container compatibility, and prepare the runtime environment. Then we’ll use VSS agent skills to turn a natural-language application request into a configured deployment, with Cosmos providing video understanding for event verification. You’ll learn how to: - Use Jetson Agent Skills to check device resources and configure the JetPack, container, and inference environment. - Use VSS 3.3 agent skills to build a visual AI application from a natural-language request. - Bring video search, alerts, and event review together in one workflow on Jetson. - Validate the deployment and add capabilities while accounting for edge memory and compute constraints. Join us live, bring your questions, and leave with a repeatable workflow for building visual AI agents on Jetson—from device setup to application deployment. Resources: VSS 3.3 Tech Blog: https://nvda.ws/4e1zxOf VSS github: https://nvda.ws/479uZBP Brev Launchable: https://nvda.ws/476yr08

    Not rated yet
  • Ask the Experts: Dory Looped Transformer Model | Nemotron Labs

    · 49 min

    Can one 4B model trade compute for accuracy at inference time, with no change to its weights? Nemotron Labs Dory 1.0, an open research model from NVIDIA, explores that question with a looped transformer that reuses a shared block of layers, taking SWE-bench Verified from 50.3 at one loop to 61.8 at four loops. In this Ask the Experts session, you’ll get direct access to the NVIDIA researchers behind it, and we’ll cover how looping works, why last-loop KV caching keeps memory flat, and what the early results suggest about trading compute for accuracy. What you’ll learn: How looped transformer models trade inference compute for accuracy, and where extra loops stop helping How last-loop KV caching cuts memory overhead for long-context workloads How native NVFP4 pre-training and the BF16, MXFP8, and NVFP4 checkpoints run on DGX Spark and Hopper and Blackwell GPUs How to evaluate loop-count tradeoffs on your own tasks Have questions about looped transformer models, loop-count tradeoffs, or what’s next for this research? Drop them live, and our guests will answer them in real time.

    Not rated yet
  • NVIDIA Dynamo in 5 Minutes: The Frontend, a Drop-In OpenAI API

    · 4 min

    Learn how NVIDIA Dynamo’s OpenAI-compatible frontend handles inference requests and routes them to reuse KV cache. In this beginner-friendly, four-minute episode of Dynamo in 5 Minutes, follow a curl request demo and see a model hot-swap in action. Learn more: https://www.nvidia.com/en-us/ai/dynamo/ Technical blog: https://developer.nvidia.com/blog/tag/nvidia-dynamo/ Get started: https://github.com/ai-dynamo/dynamo 0:00 Intro 0:33 The Problem: Changing Infrastructure 1:02 What Is the Dynamo Frontend? 1:44 Use Case: Migrating at the API Boundary 2:36 Swapping Backend Engines 3:16 Why It Matters for KV Cache 3:38 Architecture Walkthrough 4:10 Takeaway: Use an OpenAI-Compatible API to Swap Backends and Enable KV Cache Reuse

    Not rated yet
  • Where Does Self-Driving AI Get Its Training Data? 🚗

    · 36 sec

    Hot coffee, hotter take: your next AI driving lesson starts in Cosmos 3.

    Not rated yet
  • Dev Community Live: NVIDIA Developer Live Agent Gauntlet

    · 1 hr 0 min

    We had 20 developers build the agents. Now watch them put it into the unknown. The NVIDIA Developer Live Agent Gauntlet is a virtual agent competition where developers build and submit agents capable of dynamic exploration using the Model Context Protocol (MCP)—then watch them battle it out LIVE on the NVIDIA Developer livestream on Thursday, November 5 at 11:00 AM PT. At kickoff, their agents will be dropped into a Black Box MCP server filled with hidden tools, unknown functions, encrypted data, and challenges spanning Text and Image. You won't know what's inside and neither will their agents. Its job is to figure it out.

    Not rated yet
  • From Human Demonstrations to Robot Training Data With RLWRLD and Reactor | Cosmos Labs

    How do you turn a video of someone working into training data a robot can learn from? Join NVIDIA, RLWRLD, and Reactor to explore a pipeline that connects human demonstrations to robot actions, visual observations, and task instructions—and see how to run a Cosmos simulation checkpoint on Reactor. Follow the process from synchronized video capture and hand–object reconstruction to robot-specific motion retargeting and rendering in NVIDIA Isaac Sim. Explore how RLWRLD uses NVIDIA Cosmos 3 features in its RLHND hand-tracking system, then examine research demos of world-model-generated observations for garment packing and different robot configurations. Reactor will demonstrate how to run the Cosmos 3 Nano Sim Bimanual checkpoint on its platform for real-time world-model inference. Built for robotics developers and researchers, the session covers practical data-quality choices, what the current results demonstrate, and what remains to be evaluated. What You'll Learn: - Follow the pipeline from synchronized human video to aligned robot actions, observations, and task instructions. - Understand how RLHND uses NVIDIA Cosmos 3 features for hand pose, contact, and force estimation, and how reconstructed motion is retargeted to a robot. - Compare Isaac Sim-rendered observations with experimental world-model-generated observations, including deformable objects and different robot configurations. - How to run interactive world simulation models on Reactor Have questions about human demonstration data, hand tracking, retargeting, or observation synthesis? Add them to the live chat for the NVIDIA + RLWRLD + Reactor Q&A. Access more NVIDIA Cosmos developer resources and join our developer community: 📄Read the RLWRLD technical blog → https://www.rlwrld.ai/en/insight/blog/116 ⬇️ Download Cosmos on Hugging Face → https://huggingface.co/collections/nvidia/cosmos3 📚 Explore Models & Datasets on GitHub → https://github.com/nvidia/Cosmos 👥 Join the Cosmos Community → https://discord.com/invite/nvidiaomniverse

    Not rated yet
  • Train, Simulate, and Serve World Models with NVIDIA Cosmos Framework and NIM | Cosmos Labs

    · 1 hr 4 min

    NVIDIA Cosmos Framework is opening more of the training stack behind Cosmos models, giving developers reusable skills and recipes to build with Cosmos and other compatible models. Join NVIDIA experts for a practical walkthrough of how to choose a recipe, work with a compatible model, and point your coding agent to skills for training, customization, distillation, and quantization. We’ll also explore new support for offline and online reinforcement learning and what it enables for robotics developers. What You'll Learn: - How to build custom world models using Cosmos Framework - How to use offline and online RL into robotics development workflows - How to serve supported Cosmos models with Cosmos NIM - How to assess quality, GPU utilization, latency, and throughput Have questions about how to post-train and deploy NVIDIA Cosmos 3? Drop them live — the NVIDIA team will answer them in real time.

    Not rated yet
  • Dev Community Live: CUDA Rust Meetup

    · 2 hr 32 min

    Rust’s borrow checker delivers memory safety without a performance penalty, helping developers build ambitious systems with confidence. Where else can we apply that approach to provide strong correctness guarantees? Join speakers from NVIDIA, OpenAI, Zed Industries, and AWS for a livestream exploring Rust’s growing ecosystem—from CUDA and GPU computing to distributed systems. NVIDIA Senior Research Scientist Melih Elibol will introduce NVIDIA’s CUDA and Rust initiative, alongside perspectives from Dmitriy Kovalenko of OpenAI and Eric Holk of Zed Industries. Shadaj Laddad from AWS will introduce Hydro, a Rust framework that enforces strong safety guarantees for distributed systems at compile time with zero runtime overhead. Discover how Hydro enables global reasoning and seamless composition across building, testing, and deployment, and explore its latest features, including integration with formal verification. Bring your questions alongside our live audience.

    Not rated yet
  • Run Polars on a GPU with AWS | Live with cuDF Polars

    · 48 min

    You've run Polars on a GPU and scaled it across multiple GPUs. Now take it to the cloud. In this third session of the Polars livestream series, we run the same Polars code on a single-node AWS EC2 instance (NVIDIA RTX PRO 6000), reading data from S3. We start with a simple query, benchmark CPU vs. GPU, then build a visualization live. What you'll learn: - Run GPU Polars queries on a single-GPU AWS EC2 instance - How to read Parquet from S3 and switch a Polars query to the GPU engine with one line - How CPU vs. GPU Polars compare, and how to visualize results fast - Common challenges reading data from the cloud, and how to avoid them Join us as we run the code and answer questions throughout the session.

    Not rated yet
  • How McCarthy Is Building AI Agents with Palantir to Save Construction Teams Time

    · 49 sec

    McCarthy believes one AI agent could give its teams back 30,000 hours. Hear how McCarthy and Palantir are using AI to save time in preconstruction and help teams spot job site changes that could affect costs and schedules. Filmed at NVIDIA GTC Berlin.

    Not rated yet