isaacsim-performance#

The isaacsim-performance distribution (CMake group isaacsim_performance) provides standalone, Kit-independent benchmark authoring and execution through isaacsim.performance.benchmarking. Use the benchmark facade to declare operation, subsystem, and workflow benchmarks, collect named measurements, and write structured JSON results.

Operation benchmarks calibrate their batch size before measurement. Subsystem and workflow benchmarks run one unit of work per sample or frame. Every scope supports warmup, deterministic random seeds, synchronous or asynchronous targets, lifecycle hooks, validation, and optional process, GPU, and simulation metrics.

from isaacsim.performance.benchmarking import benchmark

@benchmark.operation(
    "isaacsim.example.increment",
    warmup_samples=2,
    measured_samples=10,
)
def increment(value: int = 1) -> int:
    return value + 1


result = benchmark.run(increment, save=False)
print(result["status"]["valid"])

Set ISAACSIM_BENCHMARK_OUTPUT_DIR to choose the default result root, pass an explicit path with output, or use save=False for in-memory results. Request the process, gpu, or simulation metric profile from a benchmark decorator. GPU metrics require the nvidia-ml-py runtime package and an NVIDIA GPU that is visible to NVIDIA Management Library (NVML).

Release: 7.0.0a1.

Dependencies: None.

Modules#