The latest update to NVIDIA Nsight Systems - a triage and performance analysis tool designed to track GPU workloads to their CPU origins within a system-wide view. These features help you analyze your applications’ GPU utilization, graphic and compute API activity, and OS runtime operations. This helps optimize your application to perform and scale efficiently across any quantity or size of CPUs and GPUs—from workstations to supercomputers.—is now available for download. Nsight Systems 2025.6.1 introduces several improvements aimed to enhance the profiling experience.
2025.6.1 Highlights
-
CUDA improvements
-
CUDA version compatibility update
-
System-wide CUDA Trace – Adds
--cuda-trace-scopeto select between tracing process trees or the entire system. -
CUDA Host Function Trace – Added trace support for CUDA Graph host function nodes and
cudaLaunchHostFunc(), which executes on the host and blocks the stream. -
CUDA Hardware Trace is now default – Hardware-based tracing is now default when supported, and fallback to software for various conflicts or unsupported situations. Use
--trace=cuda-swto force to software mode. -
Green Context SM Allocation Tooltip – Green context rows now show SM allocation in tooltips for better GPU resource insight.
-
-
CPU Metrics Sampling improvements - Added time-based multiplexing
-
NCCL Trace improvements
-
Detailed visibility into operations within fused GPU kernels.
-
Correlation across all events of one collective operation, properly linking API calls, runtime scheduling, and GPU operations even across different threads, processes, and CUDA graph captures.
-
-
PyTorch Trace improvements - Added forward methods and training parameters.
-
Python Sampling improvements - Better backtrace display in timeline tooltips and events view.
-
VRAM Usage Recipe
-
Analyze Windows graphics resource allocation, migration, event history, allocation callstack and perf markers.
-
Display a diff of Windows VRAM residency between graphics frames.
-
-
Plugin GUI Project Properties improvements - Added UI controls for configuration of each plugin, to be applied to the next profiling session.
-
Debuginfod Servers for Linux - Now supported along with DEBUGINFOD_URLS environment variable.
-
NVIDIA Nsight Streamer improvements now available on NGC for viewing reports on remote headless servers
-
NVIDIA Nsight Operator for Kubernetes improvements - Releasing soon on NGC
- Learn more here and apply for early access features
Nsight Systems is part of the powerful debugging and profiling NVIDIA Nsight Tools Suite. A developer can start with Nsight Systems for an overall system view and avoid picking less efficient optimizations based on assumptions and false-positive indicators.
More information
-
Subscribe to the NVIDIA Developer YouTube channel for future release highlights and feature spotlights
-
Leave a comment or question in NVIDIA Forums or visit the Nsight product page to download the latest release of Nsight Systems