NVIDIA Nsight Systems

About

NVIDIA Nsight Systems is a system-wide performance analysis tool that helps developers visualize an application's algorithms, find the largest optimization opportunities, and tune applications to scale across CPUs and GPUs. It presents system workload metrics on a unified timeline, allowing developers to investigate correlations, dependencies, activity, bottlenecks, and resource allocation. The tool traces CPU and GPU workloads, including CUDA, cuBLAS, cuDNN, TensorRT, and graphics APIs such as Vulkan, OpenGL, DirectX 11/12, DXR, and OptiX, and can sample GPU metrics such as SM utilization, PCIe throughput, NVLink, and DRAM activity.

Nsight Systems also supports multi-node profiling for data center and cluster workloads, combining network metrics with Python backtrace sampling, and integrates with JupyterLab for profiling Python code. Developers can extend it with standalone or in-process plugins to bring in additional data, and it can automatically detect slow or stuttering frames and report likely causes. It is part of the broader NVIDIA Nsight Developer Tools suite.