agentleFS
Sign inStart free
NVIDIAKnown publisherNVIDIA/omniperf/.agents/skills/benchmark-isaacsim/SKILL.md

benchmark-isaacsim

Skill34 starsChanged 7 months agoApache-2.0View on GitHub

Run Isaac Sim benchmark scripts and interpret benchmark outputs. Covers camera, SDG, scene-loading, robot, lidar/radar/sensor benchmark scripts, common parameters, output files, and benchmark-specific pitfalls. Use when the user asks to run or compare Isaac Sim benchmark results. NOT for initial bottleneck triage (use diagnose-perf), profiling capture (use profiling), trace analysis (use nsys-analyze), or applying performance fixes (use perf-tuning).

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 1 is distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Set Nucleus auth: if using Nucleus-hosted assets — see install-isaacsim skill.

    line 19
  2. Very specific:

    Do not patch Isaac Sim shutdown by default: If Tracy shutdown hangs after outputs are complete, use the scoped last-resort guidance in the …

    line 20
  3. Very specific:

    All in standalone_examples/benchmarks/.

    line 24
  4. Distinctive:

    Always verify result JSON to confirm params were applied.

    line 61
  5. Very specific:

    When comparing benchmark outputs, record whether the run used --non-headless, --viewport-updates, render products, or camera sensors.

    line 64
  6. Very specific:

    If kpis_*.json and *.tracy exist with non-zero size and the process has not exited after 2 minutes with no new output, follow the profiling …

    line 74
  7. Very specific:

    If results indicate per-camera viewport overhead, too many render products, default viewport work in headless mode, or poor multi-GPU …

    line 78

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Isaac Sim Benchmarking

Parameter references may be outdated. Always verify with ./python.sh <script> --help. For profiling details (Tracy, Nsight), see the profiling skill.

Setup

See the install-isaacsim skill for installation (pip, source build, Docker).

Before Running Any Benchmark

  1. Use a WARM run for headline FPS/frametime — see the COLD/WARM/TRACY method in the profiling skill.
  2. Set CPU governor to performance — see perf-tuning skill.
  3. Set Nucleus auth if using Nucleus-hosted assets — see install-isaacsim skill.
  4. Do not patch Isaac Sim shutdown by default. If Tracy shutdown hangs after outputs are complete, use the scoped last-resort guidance in the profiling skill.

Benchmark Scripts

All in standalone_examples/benchmarks/. Run via ./python.sh <script> [args].

ScriptWhat it measuresKey params
benchmark_camera.pyMulti-camera rendering FPS--num-cameras, --resolution W H
benchmark_sdg.pySynthetic data generation throughput--num-cameras, --annotators, --asset-count
benchmark_scene_loading.pyScene load time + FPS--env-url (required)
benchmark_robots_o3dyn.pyO3Dyn robot physics--num-robots, --physics
benchmark_robots_humanoid.pyHumanoid physics--num-robots, --physics
benchmark_robots_nova_carter.pyNova Carter (no ROS)--num-robots
benchmark_robots_nova_carter_ros2.pyNova Carter + ROS2--num-robots, --enable-3d-lidar, --enable-hawks
benchmark_robots_evobot.pyEvobot multi-phase--num-robots 1 10 20
benchmark_robots_ur10.pyUR10 manipulation--num-robots, --device
benchmark_core_world.pyCore world cloning--num-envs (no --num-frames)
benchmark_rtx_lidar.pyRTX lidar sensor--num-sensors, --lidar-type
benchmark_physx_lidar.pyPhysX lidar sensor--num-sensors
benchmark_rtx_radar.pyRTX radar sensor--num-sensors
benchmark_single_view_depth_sensor.pySingle-view depth camera--num-cameras, --resolution W H
benchmark_rtx_lidar_ros2_pcl_metadata.pyRTX lidar + ROS2 PCL (v6+)--num-sensors, --metadata
benchmark_nucleus_kpis.pyNucleus KPIs(none)

Common params: --num-frames (default 600), --num-gpus, --backend-type, --viewport-updates, --non-headless

Output control Kit args (append to any benchmark command):

--/exts/isaacsim.benchmark.services/metrics/metrics_output_folder=<output_dir>
--/log/file=<output_dir>/kit.log

Critical Gotchas

Parameter format: HYPHENS, not underscores

Isaac Sim uses hyphens in parameter names. Underscores are silently ignored by argparse:

CORRECT: --num-cameras 8 --num-gpus 1 --num-frames 600
WRONG:   --num_cameras 8  (silently uses default=1!)

Always verify result JSON to confirm params were applied.

Headless and viewport validation

When comparing benchmark outputs, record whether the run used --non-headless, --viewport-updates, render products, or camera sensors. Do not change viewport code or settings from this skill; if extra viewport work appears to affect results, hand off to perf-tuning.

JWT token expiry = silent black renders

Expired OMNI_PASS tokens cause silent asset loading failure. See install-isaacsim skill for the expiry check command.

carb.log_warn() doesn't go to stdout

Monitor progress by polling for result JSON files, not watching stdout. Kit log: ~/.nvidia-omniverse/logs/Kit/isaacsim*/kit.log or --/log/file=/tmp/kit.log.

Hung processes after results are written

If kpis_*.json and *.tracy exist with non-zero size and the process has not exited after 2 minutes with no new output, follow the profiling skill's guide-aligned shutdown handling and force-kill the app and capture processes.

Optimization Handoff

If results indicate per-camera viewport overhead, too many render products, default viewport work in headless mode, or poor multi-GPU scaling, do not apply fixes here. Summarize the benchmark evidence and use perf-tuning for the configuration changes.

Output Files

  • kpis_benchmark_<name>.json — benchmark KPIs (main results)
  • kit.log — execution log (if --/log/file= is set)
  • *.tracy — Tracy profiling trace (only with Tracy args)
  • *.nsys-rep — Nsight profiling trace (only with nsys)

Note: benchmark_result.json and startup_*.json are NOT produced by default in v5.x/v6.x. The primary results file is kpis_benchmark_<name>.json.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free