NVIDIA Developer
NVIDIA DeveloperVideo
- Format:
- YouTube channel
- New videos:
- Several times a week
- Typical length:
- about 50 minutes, not counting Shorts
- Latest video:
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.
Cosmos
Similar in subject
- AI Space
Both cover: robotics, large language models, ai agents - Practical AI
Both cover: physical ai, ai agents - ThursdAI - The top AI news from the past week
Both cover: generative ai, large language models
Similar shows
Videos
5 Ways to Tackle the AI Memory Wall
· 12 min
How can AI get more useful work from every byte of memory? In this episode of Inside the Architecture, we explore five ways NVIDIA optimizes memory use across the AI factory. Learn how model weights, KV cache, and memory tiers work together, and why efficiency depends on the full system. You’ll learn: - How NVFP4 reduces memory footprints while preserving model accuracy - How NVLink and expert parallelism distribute model weights across GPUs - How NVIDIA Dynamo manages and reuses KV cache across inference workers - How Vera accelerates tool execution and connects host memory with Rubin GPUs - How NVIDIA CMX enables context reuse across servers to reduce repeated computation No prior experience with AI memory infrastructure required. Read the full technical deep dive here: https://nvda.ws/4zuWhz6 00:00 Five Ways to Get More Useful Work From Memory 00:28 Why AI Needs Memory 01:20 NVFP4: Reducing the Memory Footprint 04:48 NVLink: Distributing Model Weights Across GPUs 06:24 NVIDIA Dynamo: Managing and Reusing KV Cache 07:28 Vera: Reducing Tool Waits and Managing Context 09:52 NVIDIA CMX: Reusing Context Across Servers 11:56 Bringing the Full System Together
Not rated yetAccelerating Structural Simulation in Abaqus with NVIDIA CUDA-X
· 33 min
Discover how Dassault Systèmes uses NVIDIA CUDA and CUDA-X libraries to accelerate structural simulation in Abaqus. This session will explore GPU acceleration across Abaqus simulation procedures, including cuDSS for sparse solving, and the development work behind these integrations. Learn how accelerated computing can help engineers shorten simulation turnaround times and evaluate complex structural behavior more efficiently.
Not rated yetInside Hermes Agent Traces with NeMo Relay | Nemotron Labs
· 51 min
Join NVIDIA and Nous Research for a live walkthrough of how agent traces turn opaque tool-use workflows into evidence developers can inspect and improve. Using Hermes Agent as a practical example, the session will show how NVIDIA NeMo Relay captures model calls, tool calls, runtime state, and handoffs inside an agent harness, then exports that telemetry to observability tools such as LangSmith or Phoenix. What you’ll learn: - How agent traces differ from LLM evals - Why NeMo Relay can support governance and optimization beyond passive tracing - What developers can take from the Hermes integration to instrument their own agent harnesses. Bring your questions about agent optimization, harnesses and traces. Our NVIDIA and Nous Research experts will be available to answer them live.
Not rated yetHow Cities Are Using AI Agents and Digital Twins
· 1 min
AI agents that answer residents’ questions. Vision agents that help reduce traffic delays. Digital twins that simulate 15 years of urban forest growth in 10 minutes. Our partners are building applications for cities with NVIDIA Nemotron, Cosmos and Omniverse, sharing computing resources across departments and municipalities. Read about the projects → https://nvda.ws/3V0KXvx
Not rated yetHow to Maximize Memory Efficiency and Reduce Costs on NVIDIA Jetson
· 1 hr 3 min
Memory optimization can unlock bigger AI models on lower-cost NVIDIA Jetson SKUs. By reducing software overhead, choosing efficient inference runtimes, and quantizing models, developers can make room for more AI - and reduce hardware costs. In this livestream, NVIDIA partners RidgeRun and Connect Tech will share demos and lessons learned on memory optimization, covering model quantization, efficient inference runtimes, and NVIDIA Jetson Agent Skills. You’ll see how large vision-language models fit into less memory and how a real robotics application moved to a more cost-efficient Jetson SKU. You’ll learn: - How much cost savings memory optimization can unlock. - How to apply quantization, runtime selection, and NVIDIA Jetson Agent Skills to fit bigger models into less memory. Join us live, bring your questions, and learn how to fit bigger models into smaller memory while reducing edge AI deployment costs
Not rated yetServing World Models: Latency vs Throughput with NVIDIA Cosmos 3 Super on Nebius | Cosmos Labs
· 57 min
Serving world models for video generation requires a different balance of latency and throughput than serving large language models. In this livestream, Nebius Physical AI team members Stewart Tong and Timothy Le examine NVIDIA Cosmos 3 Super benchmarks through vLLM-Omni on NVIDIA HGX B200 and HGX H200 systems. Their work compares four serving topologies, ranging from one eight-GPU replica to eight single-GPU replicas. See how replica count, parallelism, and concurrency affect request latency and validated video-seconds per node-hour. Learn which settings to hold constant for video generation benchmarks and how to check generated clips for file integrity and basic motion. What You'll Learn: - Why faster video requests can mean less output per node - How to measure latency and throughput with video-validation checks - When more replicas help and extra concurrency adds delay - What NVIDIA HGX B200 and HGX H200 benchmarks show about allocating GPUs - Have questions about serving NVIDIA Cosmos 3 Super for world-model video generation? Drop them in live — the NVIDIA and Nebius teams will address them during the discussion. Learn more about NVIDIA Cosmos 3 serving and reproduce the benchmark: 📄 Read the technical blog → https://nebius.com/blog/posts/serving-world-models-latency-node-throughput 🧪 Reproduce the benchmarks → https://github.com/nebius/nebius-physical-ai 📚 Explore the NVIDIA Cosmos platform → https://github.com/nvidia/Cosmos 👥 Join the NVIDIA Omniverse community → https://discord.com/invite/nvidiaomniverse
Not rated yetMeet NVIDIA Nemotron 3.5 Transcribe 🎙️
· 53 sec
NVIDIA Nemotron 3.5 Transcribe is an open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint. Download from @HuggingFace: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B
Not rated yetCUDA Live: Introduction to CUDA Rust
· 39 min
CUDA Rust brings GPU kernel programming to Rust, with two tracks: cuda-oxide for SIMT kernels and cutile-rs for tile-based kernels. In this livestream, Andy Terrel and the CUDA Rust team write, build and run GPU kernels in Rust live, explain when to use each track, and answer your questions. What you'll learn: How the two CUDA Rust tracks differ: cuda-oxide for SIMT kernels and cutile-rs for tile-based kernels How to write, build and run your first GPU kernel in Rust Which track fits your project, and how to share feedback with the CUDA Rust team Join us live and bring your questions for the CUDA team.
Not rated yetPost-Training Nemotron for Europe | Nemotron Labs
· 47 min
At GTC Berlin, NVIDIA showed what’s possible when you post-train Nemotron for a specific region, language, and cultural context, and this tutorial brings that work to your terminal. We’ll walk through the full post-training pipeline using Nemotron Persona open datasets: from defining target behavior and generating synthetic conversations to fine-tuning with LoRA and evaluating the result. You’ll see exactly how this workflow produces models that handle regional terminology, cultural nuance, and domain-specific response patterns that general-purpose models miss. What you’ll learn: - How to use Nemotron Persona datasets to generate synthetic training conversations for your target language and region - How to fine-tune Nemotron with LoRA for domain- and locale-specific behavior - How to merge the adapter and serve your adapted model - How to evaluate the adapted model against the base to measure improvement Working on localizing a model or adapting Nemotron for your region? Bring your questions, and our speakers will walk through the workflow live and take them at the end.
Not rated yetBuild AI Models for Crash Analysis with NVIDIA PhysicsNeMo
· 5 min
Learn how to build AI models that use crash simulation data to predict how vehicle structures deform, helping engineers evaluate more designs faster. This video walks through the NVIDIA PhysicsNeMo workflow, from preparing meshes, design variables and time-series data to training and evaluating an AI surrogate model. Compare one-shot, time-conditional and autoregressive prediction approaches, and see how an optional guardrail can flag unfamiliar inputs for further review. Get started with the PhysicsNeMo crash training recipe and companion bumper-beam notebooks covering data preparation, model architecture, training, inference and validation. High-fidelity simulations and physical testing remain essential for validation and final engineering decisions. Learn how to build AI surrogate models for structural and automotive crash analysis: https://nvidia.github.io/physicsnemo/blog/2026/03/17/structural-mechanics/ Explore the NVIDIA PhysicsNeMo crash training recipe: https://github.com/NVIDIA/physicsnemo/tree/main/examples/structural_mechanics/crash Work through the crash-analysis notebooks: https://github.com/openhackathons-org/End-to-End-AI-for-Science/tree/main/workspace/python/jupyter_notebook/Crash Learn Transolver and GeoTransolver from theory through training: https://github.com/openhackathons-org/End-to-End-AI-for-Science/tree/main/workspace/python/jupyter_notebook/Transolver Chapters: 00:00 - Introduction & AI Surrogates for Crash Analysis 00:57 - Dataset Creation, Validation & Guardrails 01:36 - Converting Simulation Data to AI Formats 02:15 - Prediction Methods: One-Shot, Time-Conditioned & Autoregressive 02:53 - PhysicsNeMo Crash Recipe Overview 03:50 - Overview of Educational Notebooks 04:10 - Notebook 0: Data Preparation & Inspection 04:45 - Notebook 1: Neural Transolver Architecture & GALE Attention 05:20 - Notebook 2: Training, Inference & Model Evaluation
Not rated yetHow to Post-Train NVIDIA Nemotron for French
· 4 min
Post-train NVIDIA Nemotron 3.5 Lightning for your users’ language and local context. This five-minute tutorial walks through a French support example using Nemotron-Personas datasets, synthetic conversations, and LoRA fine-tuning with NeMo AutoModel. Learn how to define target behavior, generate and check training data, fine-tune your model, and compare its responses with the base model—then adapt the workflow to other languages. Nemotron code and recipes: https://github.com/NVIDIA-NeMo/Nemotron French persona dataset: https://huggingface.co/datasets/nvidia/Nemotron-Personas-France NeMo AutoModel: https://github.com/NVIDIA-NeMo/Automodel Tech blog and exact model download: https://developer.nvidia.com/blog/fine-tuning-nvidia-nemotron-to-know-local-rules-not-just-local-languages Chapters (from supplied transcript; confirm against final video): 00:00 Post-Training for Language and Local Context 00:37 French Support: Base vs. Post-Trained Model 01:13 Personas, Synthetic Data, and LoRA 01:44 Define Target Behavior 01:57 Generate Synthetic Conversations 02:25 Audit and Split Training Data 02:39 Fine-Tune With LoRA 03:05 Serve and Evaluate Your Model 03:48 Further Training and Other Languages 04:17 Get Started
Not rated yetFaster LLM Inference on Jetson With Speculative Decoding
· 36 sec
How does speculative decoding speed up LLM inference on NVIDIA Jetson? Watch Qwen3.5 9B NVFP4 generate responses with speculative decoding enabled and disabled. By drafting multiple tokens for the target model to verify together, speculative decoding can reduce sequential generation steps and deliver faster responses. Results depend on the model, hardware, and workload. ➡️ Learn more: https://developer.nvidia.com/blog/an-introduction-to-speculative-decoding-for-reducing-latency-in-ai-inference 📝 Tech blog: https://developer.nvidia.com/blog/frontier-reasoning-reaches-the-edge-how-to-deploy-and-optimize-models-on-nvidia-jetson ➡️ Explore models and turotials: https://www.jetson-ai-lab.com/
Not rated yetHow 6 Companies Are Putting Physical AI to Work With NVIDIA Cosmos
· 1 min
See how Encord, Fogsphere, LTX, Mercedes-Benz, Porsche and Siemens are putting physical AI to work with NVIDIA Cosmos, from developing world models and generating training data to improving autonomous driving and factory operations. Learn more about the new Cosmos tools for training, testing and deploying world models: https://nvda.ws/4zBcceU
Not rated yetHow AI Simulates Thousands of Tropical Cyclone Seasons
· 1 min
AI-powered tropical cyclone simulation is changing how the world models hurricane risk. In this video, AXA and NVIDIA show how FourCastNet 3 on NVIDIA Earth-2 generates thousands of synthetic cyclone seasons in the time it would take a physics-based model to run just one — giving insurers and risk teams an unprecedented view of storm activity across every major ocean basin. The historical record covers roughly 45 years of observed storms. That sounds like a lot, until you need to price risk for the next 50. A single AI ensemble member runs in under 10 minutes on one H100 GPU, making it practical to simulate at a scale that physics-based models simply cannot match. The result: a dense, global picture of where cyclones travel, how often they co-occur in the same season, and what the historical record was never long enough to show. This work was presented at GTC Berlin, October 2026, as part of an ongoing AXA and NVIDIA collaboration to apply machine learning to catastrophe risk modeling. Read the announcement: https://blogs.nvidia.com/blog/europe-earth-2-open-models-weather-climate-prediction Learn more about NVIDIA Earth-2: https://www.nvidia.com/en-us/high-performance-computing/earth-2/
Not rated yetGTC Dev Community Berlin Recap: From Open Models to Physical AI
· 3 hr 3 min
Explore what developers can build with the ideas and technology showcased at GTC Berlin. Join NVIDIA experts, model builders, and industry practitioners for conversations and demos spanning customizable open models and physical AI. Discover how these tools connect from experimenting on your desktop to powering industrial workflows and intelligent robots. Hear how builders and industry practitioners approach real use cases, discover tools and ideas for your next project, and join the conversation through live chat and audience Q&A. This stream brings the GTC experience directly to you.
Not rated yetGTC Berlin After Hours: Keynote Reactions & What’s Next for Developers
· 1 hr 3 min
The keynote is over. Let’s talk about what it means for developers. Join NVIDIA experts and the community for an after-hours conversation from GTC Berlin, featuring first reactions, a look at the day’s announcements, and perspectives on open models and agent safety. Hear what’s generating excitement across the event, test your knowledge in an interactive quiz, and enjoy a few community moments along the way. Bring your questions and help shape the discussion about what comes next.
Not rated yetAsk the Experts: Nemotron 3.5 Super VL | Nemotron Labs
· 56 min
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 researchers who built it: the architecture choices, training techniques, and design principles behind a model optimized to run across DGX Spark to the data center, with open weights and recipes you can customize for your domain. What you’ll learn: - 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 Nemotron 3.5 Super VL is optimized to run locally on hardware like DGX Spark without sacrificing accuracy or speed - How to integrate Nemotron 3.5 Super VL into multi-model agentic pipelines Have questions about training, architecture, or running Nemotron 3.5 Super VL in your stack? Ask them live, and our researchers will answer them directly.
Not rated yetHow the Nordics Are Building with AI - Featuring Legora, Corti & Root Bio | Build-a-Claw Stockholm
· 24 min
What does it take to build AI agents that can move beyond answering questions and actually do meaningful work? Recorded live at NVIDIA Build-a-Claw Stockholm on October 15, 2026, this panel explores how agentic AI is already being applied across legal tech, healthcare, biotech and scientific discovery — and what teams are learning as they deploy it in the real world. The conversation covers scaling quickly without sacrificing reliability, AI agents in regulated industries, open and sovereign AI models, the future of legal and clinical workflows, automating biological discovery, and the role the Nordics can play in the next generation of AI. Moderator: Sofia Lindelöw Panelists: • Jacob Lauritzen — Legora • Lars Maaløe — Corti • Martin Blomberg — Root Bio
Not rated yetRoad to GTC Berlin: Europe’s AI Builders in the Spotlight
· 58 min
Meet the developers shaping Europe’s AI future. Join us for highlights from developer events across Barcelona, Paris, London, Stockholm, and Berlin - from Build-a-Claw demos and community gatherings to conversations with NVIDIA experts and local builders. Explore how developers are working with AI agents, open models, and physical AI, hear their questions and experiences, and get a preview of the themes heading into GTC Berlin. Bring your questions and join the conversation live.
Not rated yetYour Spark, Your Story: Selena Williams | NVIDIA RTX Spark
· 1 min
A little curiosity can take you a long way. For Selena, it led to building robots, exploring AI for well-being, and finding new ways to create with NVIDIA RTX Spark. Learn more about NVIDIA RTX Spark: https://www.nvidia.com/en-us/products/rtx-spark/
Not rated yet