NVIDIA/omniperf

NVIDIA/omniperf

Description: Tracking Isaac Lab performance across GPUs and simulation backends.

Language: Python

License: Apache-2.0

Stars: 32

Forks: 3

Open issues: 0

Created: 2026-03-06T23:57:31Z

Pushed: 2026-05-14T02:08:49Z

Default branch: main

Fork: no

Archived: no

README:

Omniperf — Performance AI Skills and Dashboards for Omniverse, Isaac Sim, and Isaac Lab

Omniperf bundles two things for performance work across Omniverse, Isaac Sim, and Isaac Lab:

1. Performance AI Skills — a library of Claude Code agent skills that install, benchmark, profile, and diagnose Kit-based applications. See [Agent Skills](#agent-skills) below. 2. Performance Dashboards — a static site that tracks benchmark results (FPS, GPU utilization, memory, startup times) across GPUs and historical CI runs.

Live dashboard:

Overview

The dashboards are a lightweight, zero-dependency static site driven by JSON benchmark data generated by internal CI/CD pipelines and committed to this repo. The AI skills live under [.agents/skills/](.agents/skills/) and encode the install, benchmark, and profile-analysis workflows used to produce and interpret that data.

Dashboard features

• Multi-GPU comparison — switch between GPU datasets (e.g. NVIDIA L40, RTX PRO 6000)
• Overview tab — sparkline grid of all benchmarks grouped by workflow (Non-RL, RL Games, RSL RL)
• Detail tab — full time-series charts, per-run summaries, and an all-benchmarks overview table
• Search & filter — autocomplete search to quickly find specific benchmarks
• Metric selection — explore runtime metrics (FPS, GPU/CPU utilization, memory) and startup metrics

Getting Started

Local Development

git clone https://github.com/NVIDIA/omniperf.git
cd omniperf
./serve.sh
# Open http://localhost:8765

Requires Python 3 (for the built-in HTTP server). No build step or dependencies needed.

> Note: The default clone skips the benchmark preview images (~1.5 GB) to keep clones lightweight. To also download them, run git lfs pull --exclude="" after cloning.

Data Format

Benchmark data lives in docs/data/. The manifest.json file lists available GPU datasets:

{
"generated_at": "2026-03-04T10:47:26Z",
"datasets": [
{
"file": "ada_l40.json",
"gpu_display_name": "NVIDIA L40",
"total_runs": 228
}
]
}

Each GPU data file contains an array of benchmark runs with commit metadata, runtime metrics, and startup metrics.

Adding New Data

1. Generate the benchmark JSON using the omniperf tooling 2. Place the JSON file in docs/data/ 3. Update docs/data/manifest.json to reference the new file 4. Commit and push — GitHub Pages will automatically deploy

Deployment

The site is deployed automatically via GitHub Actions to GitHub Pages on every push to main. The workflow:

1. Checks out the repo (with LFS) 2. Uploads docs/ as a Pages artifact 3. Deploys to GitHub Pages

To enable GitHub Pages for a fresh clone: 1. Go to Settings → Pages in the GitHub repo 2. Under Source, select GitHub Actions

Requirements

• Data files are tracked with Git LFS — install it before cloning
• Chart.js 4.4.7 is loaded from CDN (no npm install needed)
• Python 3 for local development server

Agent Skills

This repo ships a set of [Claude Code agent skills](.agents/skills/) for working with Omniverse, Isaac Sim, Isaac Lab, and their profiling tools. They encode install steps, benchmark recipes, and profile-analysis workflows validated from live testing.

• [install-isaacsim](.agents/skills/install-isaacsim/SKILL.md) — install Isaac Sim via pip or source build
• [install-isaaclab](.agents/skills/install-isaaclab/SKILL.md) — install Isaac Lab for Isaac Sim-backed or kit-less/Newton workflows
• [install-profilers](.agents/skills/install-profilers/SKILL.md) — set up Nsight Systems, Tracy, and related tooling
• [benchmark-isaacsim](.agents/skills/benchmark-isaacsim/SKILL.md) — run Isaac Sim benchmarks
• [benchmark-isaaclab](.agents/skills/benchmark-isaaclab/SKILL.md) — run Isaac Lab RL and environment benchmarks
• [profiling](.agents/skills/profiling/SKILL.md) — capture traces with Tracy and Nsight Systems
• [profiling-api](.agents/skills/profiling-api/SKILL.md) — add profiling zones, metrics, and annotations to Kit code
• [nsys-analyze](.agents/skills/nsys-analyze/SKILL.md) — analyze Kit-based .nsys-rep profiles and compare versions
• [tracy-memory](.agents/skills/tracy-memory/SKILL.md) — profile CPU and GPU memory allocations in Tracy
• [nvtx-python](.agents/skills/nvtx-python/SKILL.md) — trace Python functions with NVTX outside Kit/Carbonite
• [diagnose-perf](.agents/skills/diagnose-perf/SKILL.md) — first-responder triage for slow FPS, stutter, or latency
• [perf-tuning](.agents/skills/perf-tuning/SKILL.md) — apply guide-backed fixes for known performance bottlenecks
• [kit-app-streaming-debug](.agents/skills/kit-app-streaming-debug/SKILL.md) — debug Kit app livestream bottlenecks using StreamSDK logs and browser WebRTC stats

Security

• Vulnerability disclosure: See SECURITY.md

Contributing

See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines, including the Developer Certificate of Origin (DCO) requirement.

License

This project is licensed under the [Apache License 2.0](LICENSE).