About
NVIDIA Nsight Compute is an interactive profiler for CUDA and NVIDIA OptiX that provides detailed performance metrics and API debugging through both a user interface and a command-line tool. It offers guided analysis and data-driven comparisons, and results can be post-processed and analyzed within a user's own workflows. It is also available as part of the CUDA Toolkit.
The tool surfaces hardware performance metrics in an understandable form, with built-in guidance from NVIDIA's rule set identifying common performance limiters and suggesting optimizations. Its report pages cover GPU utilization, memory workload with heatmap visualizations of the memory pipeline, and warp stall sampling. Metrics can be correlated down to individual source lines, connecting SASS with PTX and higher-level code such as CUDA C/C++, Fortran, OpenACC, and Python. Interactive profiling supports stepping through API calls and exploring CUDA graphs.
Nsight Compute also supports customization and collaboration, including custom metric collection, baseline comparisons across GPU architectures, sharing reports with colleagues, and profiling through the GUI or CLI, locally or remotely. Python developers can automate analysis with the NVRules API. It is part of the NVIDIA Nsight Developer Tools suite, alongside tools such as Nsight Graphics.
What's new
All release notesHomebrew cask nvidia-nsight-compute 2026.3.1.2 (sha256 fba5201ddd8a42294dba708e3595c2599d603fb0430e82fcdb2b7cf5680b564b).