NVIDIA at SIGGRAPH 2026

First Offering:

Sunday, July 19
2–2:25 p.m. PT
NVIDIA Lab Room, Room 502B

Second Offering:

Monday, July 20
9–9:25 a.m.
NVIDIA Lab Room, Room 502B

This presentation offers a focused update on the state of NVIDIA's graphics developer tools and a look at how agentic AI workflows are shaping their future direction. Covering recent advances in Nsight Graphics, Nsight Systems, and Nsight Aftermath, the session highlights key improvements to debugging, profiling, and crash analysis capabilities alongside an exploration of where AI-driven automation is heading next.

Attendees will come away with a grounded understanding of the current tooling landscape and practical insight into the agentic AI features being built for graphics developers. This presentation is paired with two hands-on labs where participants will directly use and explore AI-assisted workflows.

First Offering:

Sunday, July 19
2:30–4 p.m. PT
NVIDIA Lab Room, Room 502B

Second Offering:

Monday, July 20
9:30–11 a.m.
NVIDIA Lab Room, Room 502B

Modern graphics applications are getting increasingly difficult to debug. Ray Tracing, advanced rendering pipelines with neural rendering, GPU crashes, and performance bottlenecks often require developers to navigate large amounts of data before identifying the true source of a problem. At the same time, AI-powered tools are beginning to transform how developers investigate and understand complex systems.

In this hands-on lab, participants will enter the Hall of Mysteries, an interactive graphics debugging experience built around a series of realistic rendering and performance challenges. Using a preview of the in-development Nsight Graphics MCP (Model Context Protocol) server, participants will study a Vulkan application to investigate visual artifacts, analyze renderer behavior, diagnose GPU crashes, and optimize performance using professional graphics debugging workflows. Rather than following a scripted tutorial or traditional use of the tooling, participants will gather evidence, formulate hypotheses, inspect captures and traces, and verify conclusions through AI-guided investigation. Along the way, they will learn how AI-assisted workflows can accelerate debugging while preserving the reasoning and verification skills required to solve real graphics problems.

By the end of the lab, participants will have solved multiple graphics debugging challenges and gained practical experience using AI-assisted workflows to investigate and resolve issues in modern graphics applications.

First Offering:

Sunday, July 19
4:15–5:15 p.m. PT
NVIDIA Lab Room, Room 502B

Second Offering: 

Monday, July 20
11:15 a.m.–12:15 p.m.
NVIDIA Lab Room, Room 502B

This training lab offers graphics developers a look at the agentic future being built into Nsight Systems, with a focus on holistic analysis of CPU and GPU performance. Participants will work through hands-on scenarios where AI agents assist with performance analysis. The session bridges the gap between emerging model capabilities and the practical realities of real-time and offline graphics pipelines. Developers at all levels will gain actionable insight into how agentic AI workflows can be adopted today and what innovations are on the horizon.

Monday, July 20
1–3 p.m. PT
NVIDIA Lab Room, Room 502B

In this lab, we’ll work hands-on across three core capabilities of NVIDIA Cosmos™ 3: reasoning, world generation, and action prediction. We’ll use Cosmos 3 Nano Reasoner to caption real robot video, generate augmentation prompts, and validate synthetic footage using chain-of-thought reasoning. We’ll then use Cosmos 3 and Cosmos-Transfer2.5 to synthesize and transform photorealistic video from those prompts, scaling training data without new recording sessions.

Finally, we’ll apply Cosmos 3 in different action modes to translate scene understanding into embodied control commands, completing the perceive → reason → act loop.

Tuesday, July 21
10:15–11:45 a.m. PT
Concourse Hall

This hands-on course introduces Slang, an open source, open governance shading language hosted by Khronos that simplifies graphics development across platforms. Designed to tackle the growing complexity of shader code, Slang offers modern programming constructs while maintaining top performance on current GPUs.

Tuesday, July 21
12–1:30 p.m. PT
Concourse Hall

Learn to build GPU ML inference pipelines with Slang. Use a lightweight, open set of pre-written kernels to implement an MNIST CNN from scratch: load SafeTensors weights, run inference, and apply kernel fusion for better performance. Hands-on, cross-platform, and highly customizable—ideal for games, tools, and research.

Wednesday, July 22
9–10:30 a.m. PT
NVIDIA Lab Room, Room 502B

Welcome to the first course in the NVIDIA Physical AI Agent Bootcamp. 

Physical AI systems need visual worlds that can be rendered, inspected, queried, and trusted before they are used for simulation, robotics testing, or training. This hands-on course introduces that foundation by guiding attendees through building an OpenUSD-based RTX™ viewport with NVIDIA Omniverse™ libraries and AI-agent-assisted development.

Starting from a minimal Python renderer, attendees will create an interactive app that loads USD scenes, displays RTX output, exposes camera views, queries scene attributes, and writes image results. They will also learn a repeatable agent workflow: Define the goal, provide context, name the relevant skills, review the generated diff, and run only after the code is understood.

By the end, participants will have a working RTX viewport application and a practical method for using AI agents to build reliable graphics and simulation tools for physical AI workflows.

Wednesday, July 22
10:15–11:45 a.m. PT
Concourse Hall

Hands-on course on Newton, a GPU-accelerated physics engine, and its integration with Isaac™ Lab for robot learning. Covers Newton’s architecture and import pipeline, solver and contact tuning, articulation and sensors, then training and evaluating policies in Isaac Lab with Newton as the physics backend.

Wednesday, July 22
10:45 a.m.–12:15 p.m. PT
NVIDIA Lab Room, Room 502B

Physics is at the heart of building reliable physical AI systems, and this course in the NVIDIA Physical AI Agent Bootcamp explores how to move from visual inspection into simulation readiness.

Workflows in physical AI need assets that are structured, validated, and physically meaningful. Building on the prior OpenUSD and RTX viewport lab, this hands-on course moves from visual inspection into simulation readiness.

Attendees will validate OpenUSD assets for simulation-ready (SimReady) workflows, convert CAD and Gaussian splat content to USD, create a physics stage, and assemble environments such as stairs, ramps, or other test layouts. They will enable physics, make scene objects interactable, and run simulations to see how assets behave under gravity and user input.

By the end, attendees will have a runnable physics scene with validated assets, converted USD content, and an interactive environment ready for robotics testing or physical AI workflows.

Wednesday, July 22
12–1:30 p.m. PT
Concourse Hall

Hands-on course on preparing robot USD assets for simulation, covering why physics tuning is foundational to robot learning, inspecting and configuring colliders and joints with OpenUSD/PhysX™, and tuning control gains for stable, realistic motion. Participants work with a robot hand asset and apply best practices for simulation-ready authoring.

Wednesday, July 22
1:45–3:15 p.m. PT
NVIDIA Lab Room, Room 502B

Blender is where many 3D ideas first take shape, and this course in the NVIDIA Physical AI Agent Bootcamp shows how to extend that familiar workspace with Omniverse libraries instead of moving the work somewhere else. Attendees will use AI agents to scaffold a Blender integration that connects ovrtx, ovphysx, and ovstage into a live creative workflow.

The session walks through bringing GPU-accelerated path-traced rendering into Blender’s viewport. Along the way, attendees will see how Blender’s scene graph can connect to Omniverse-powered rendering and simulation services.

By the end, participants will have an RTX-rendered, physics-enabled scene running inside Blender, plus a clearer model for how composable Omniverse libraries can extend existing tools used by artists, technical directors, and simulation teams.

Wednesday, July 22
3:30–5 p.m. PT
NVIDIA Lab Room, Room 502B

This hands-on course in the NVIDIA Physical AI Agent Bootcamp turns an empty USD stage into a working Isaac Sim™ robot scene, with AI agents acting as simulation setup partners. Attendees will import a robot manipulator from URDF, compose the scene using prior lab assets or user-selected environments, and bring SimReady content into Isaac for testing.

The session focuses on the practical path from “nothing is in the scene” to “the robot picked up the cube.” Participants will configure the robot, prepare the object, validate physics behavior, tune the scene, and execute a basic pick-and-place task. Agent skills handle repeatable setup steps so attendees can focus on scene intent, simulation checks, and task behavior. 

By the end, attendees will have a composed Isaac Sim scene with a physics-validated robot, interactive assets, and a successful pick-and-place task that connects OpenUSD scene building to robotics simulation.

Thursday, July 23
10:15–11:45 a.m. PT
Concourse Hall

We introduce a novel capture-to-simulation pipeline for in-the-wild 3D Gaussian Splat scenes, showing our research-based solutions for segmenting objects, predicting their volumetric mechanical properties using a recent learning-based method, and mixed splat-mesh physics simulations, all supported by recent features in the Kaolin Library.

Thursday, July 23
12–1:30 p.m. PT
Concourse Hall

Learn how to reconstruct a large scene for robotics testing using NVIDIA Omniverse NuRec Gaussian-based reconstruction technologies to perform multi-GPU training, object segmentation, and extraction. This lab will walk through core concepts with a step-by-step workflow for data capture, reconstruction, and object level-integration for robotics simulation in NVIDIA Isaac Lab.