Summary

NVIDIA Nsight Systems 2019.4 is a powerful tool designed to help developers optimize and scale software across CPUs and GPUs. This release introduces new data sources, improved visual data navigation, expanded CLI capabilities, and enhanced export features. Key highlights include support for MPI, Linux ftrace, Windows ETW, and DirectML, along with a new color scheme for CUDA memory transfers and the ability to produce multiple reports with a single application run.

Understanding Nsight Systems 2019.4

Nsight Systems is an indispensable tool for developers aiming to tune and scale software across CPUs and GPUs. The 2019.4 release brings significant enhancements to data collection, exploration, and control, making it a versatile tool for various markets, including high-performance computing and visual effects.

New Data Sources

The 2019.4 release introduces several new data sources that enhance the tool’s capabilities:

  • MPI: Support for Message Passing Interface (MPI) allows developers to analyze parallel computing applications more effectively.
  • Linux ftrace: Integration with Linux ftrace provides detailed system-wide tracing capabilities.
  • Windows ETW: Support for Windows Event Tracing (ETW) enables comprehensive analysis of system events.
  • DirectML: Inclusion of DirectML allows for the analysis of machine learning workloads.

Improved Visual Data Navigation

The GUI timeline events can now be viewed as a data table, offering a more detailed and organized way to analyze data. Additionally, CUDA memory transfers have been updated with a new color scheme, making it easier to identify their type and direction at a glance.

Expanded CLI Capabilities

The command-line interface (CLI) has been enhanced with new options for waiting, terminating processes, and capture range triggers. This includes the ability to produce multiple reports with a single application run using the interactive command subset (launch, start, and stop).

Enhanced Export Features

The export capabilities have been expanded, allowing for more comprehensive data export and statistics. This includes streaming export for larger report files and reduced RAM utilization.

Key Features and Improvements

General Enhancements

  • Event Table: An alternative view of timeline data provides a more detailed and organized way to analyze events.
  • CUDA Memory Transfer Color Scheme: A new color scheme helps identify the type and direction of CUDA memory transfers more easily.
  • Expanded Export Capabilities: Enhanced export features include streaming export for larger report files and reduced RAM utilization.

Windows-Specific Features

  • Manifest-Based ETW Support: Enhanced support for Windows ETW provides more detailed system event tracing.
  • Start & Stop Capture via Hotkey: Users can now start and stop captures using hotkeys for more convenient profiling.
  • SLI Log: Support for SLI (Scalable Link Interface) logging provides additional insights into multi-GPU configurations.

Linux Workstation and Server Features

  • Ftrace Collection: Integration with Linux ftrace allows for detailed system-wide tracing.
  • CUDA API Backtrace Option: Enhanced support for CUDA API backtracing provides more detailed insights into CUDA applications.
  • MPI API Trace: Support for MPI API tracing enables comprehensive analysis of parallel computing applications.
  • CLI Enhancements: Expanded CLI capabilities include export on stop, kill signal on stop, print statistics on stop, wait for all processes, and return app exit codes when using the “profile” command.

Key Takeaways

  • New Data Sources: MPI, Linux ftrace, Windows ETW, and DirectML enhance the tool’s capabilities.
  • Improved Visual Data Navigation: GUI timeline events can be viewed as a data table, and CUDA memory transfers have a new color scheme.
  • Expanded CLI Capabilities: New options for waiting, terminating processes, and capture range triggers.
  • Enhanced Export Features: Streaming export for larger report files and reduced RAM utilization.

By leveraging these features, developers can gain deeper insights into their applications and make informed decisions to optimize performance.

Conclusion

Nsight Systems 2019.4 is a powerful tool for developers looking to optimize and scale software across CPUs and GPUs. With its new data sources, improved visual data navigation, expanded CLI capabilities, and enhanced export features, it offers a comprehensive solution for performance analysis and optimization. Whether you’re working in high-performance computing, visual effects, or other fields, Nsight Systems 2019.4 provides the tools you need to achieve optimal performance.