Nsight Python

Type: Technology Tags: NVIDIA, Nsight, Python, Profiling, CUDA, Performance, Kernel Analysis Related: Nsight-Developer-Tools, Nsight-JupyterLab-Extension, CUDA-Python, Nsight-Compute, Nsight-Systems, CUPTI-Python, PyTorch, CuPy, NVIDIA-CUDA Sources: https://docs.nvidia.com/nsight-python/index.html, https://docs.nvidia.com/nsight-python/overview/quickstart.html, https://developer.nvidia.com/nsight-python, https://nvidia.github.io/cuda-python/latest/ Last Updated: 2026-04-29

Summary

Nsight Python is NVIDIA’s Python kernel profiling interface for automating performance analysis across multiple kernel configurations with NVIDIA Nsight tools. It provides Python APIs such as nsight.analyze and nsight.annotate so Python GPU workloads can collect profiling results without leaving the Python workflow.

Detail

Purpose

Python GPU developers often need to compare kernel implementations, matrix sizes, parameters, or library calls repeatedly. Nsight Python wraps Nsight profiling workflows in a Python-first interface so users can decorate benchmark functions, annotate regions, collect metrics, and analyze results programmatically.

Key capabilities

• Python kernel profiling interface built around NVIDIA Nsight tools.
• nsight.analyze APIs for running profiling analysis.
• nsight.annotate context/decorator patterns for marking the kernel region to profile.
• Quickstart flow that decorates a Python function and returns results as a pandas DataFrame.
• Documentation sections for core concepts, architecture, installation, API reference, collection, extraction, transformation, visualization, and a Nsight Compute collector.
• Developer page positioning around decorators, context manager APIs, configuration sweeps, report generation, and plotting/visualization.

NVIDIA context

Nsight Python fills a different niche from Nsight-Compute and Nsight-Systems. Those tools are the profilers; Nsight Python is a Python automation layer that makes them easier to drive inside CUDA-Python, PyTorch, CuPy, and custom kernel workflows.

Connections

• CUDA-Python - current CUDA Python hub lists Nsight Python.
• Nsight-Compute - Nsight Python includes a collector for Nsight Compute-style kernel metrics.
• Nsight-Systems - adjacent Nsight profiler for system-wide timelines.
• Nsight-JupyterLab-Extension - adjacent notebook workflow for profiling selected JupyterLab cells.
• CUPTI-Python - lower-level Python profiling/trace interface around CUPTI concepts.
• PyTorch and CuPy - common Python GPU workloads that Nsight Python can profile.
• NVIDIA-CUDA - parent CUDA platform and performance tooling ecosystem.

Source Excerpts

• NVIDIA describes Nsight Python as a Python kernel profiling interface that automates performance analysis across multiple kernel configurations using NVIDIA Nsight tools.
• The quickstart shows adding an nsight.analyze.kernel decorator and using nsight.annotate() around the kernel region to profile.