Nsight Systems is available for multiple targets and multiple host OSs. To choose the right package, first consider the target system to be analyzed.
For Tegra target systems, select Nsight Systems for Tegra available as part of NVIDIA JetPack SDK and NVIDIA CodeWorks SDK.
For x86_64 or IBM Power target systems, select from the target packages from Nsight Systems for Workstations, available from https://developer.nvidia.com/nsight-systems. This web release will always contain the latest and greatest Nsight Systems features.
The x86_64 target version of Nsight Systems is also available in the CUDA ToolKit
Each package is limited to one architecture. For example, Tegra packages do not contain support for profiling x86 targets, and x86 packages do not contain support for profiling Tegra targets.
After choosing an appropriate target version, select the package corresponding to the host OS, the OS on the system where results will be viewed. These packages are in the form of common installer types: .msi for Windows; .run, .rpm, and .deb for x86 Linux; .deb and .rpm for Linux on IBM Power; and .dmg for the MacOS installer.
Note: the IBM Power package does not have a GUI for visualization of the result. If you wish to visualize your result, please download and install the GUI available for MacOS, x86_64 Linux and Windows systems.
Tegra packages
Windows host - Install .msi on Windows machine. Enables remote access to Tegra device for profiling.
Linux host - Install .run on Linux system. Enables remote access to Tegra device for profiling.
MacOS host - Install .dmg on MacOS machine. Enables remote access to Tegra device for profiling.
x86_64 packages
Windows Host - Install .msi on Windows machine. Enables remote access to Linux x86_64 device for profiling.
Linux host - Install .run, .rpm, or .deb on Linux system. Enables remote access to Linux x86_64 device for profiling or running collection on localhost.
Linux CLI only - The Linux CLI is shipped in all x86 packages, but if you just want the CLI, we have a package for that. Install .deb on Linux system. Enables only CLI collection, report can be imported or opened in x86_64 host.
MacOS host - Install .dmg on MacOS machine. Enables remote access to Linux x86_64 device for profiling.
IBM Power packages
Copy the appropriate file to your host system in a directory where you have write and execute permissions. Run the install file, accept the EULA, and Nsight Systems will install on your system.
The installation will create a Host directory for this host and a Target directory for each target this Nsight Systems package supports.
All binaries needed to collect data on a target device will be installed on the target by the host on first connection to the device. There is no need to install the package on the target device.
If installing from the CUDA Toolkit - See CUDA Toolkit documentation
Linux x86_64 targets can be profiled using the CLI. IBM Power targets can only be profiled using the CLI. The CLI is especially helpful when scripts are used to run unattended collections or when access to the target system via ssh is not possible. In particular, this can be used to enable collection in a Docker container.
The CLI can be found in the Target directory of the Nsight Systems installation. Users who want to install the CLI as a standalone tool can do so by copying the files within the Target directory to the location of their choice.
If you wish to run the CLI without root (recommended mode) you will want to install in a directory where you have full access.
To get started using the CLI, run nsys --help
for a list of options or
see Profiling Applications on Linux Devices from the CLI for full documentation.
Depending on your OS, Nsight Systems will have installed an icon on your host desktop that you can use to launch the GUI. To launch the GUI directly, run the nsight-sys executable in the Host subdirectory of your installation.
Nsight Systems is installed by the CUDA Toolkit installer in /usr/local/cuda-[version]/NsightSystems-[version]. To launch the GUI, run the nsight-sys executable in the Host subdirectory there.
In a future version of the CUDA Toolkit, this installation will be moved under the CUDA/bin directory.
For more information on next steps in the GUI, see the section on profiling applications for your target in the Nsight Systems documentation.
Copyright (c) 2012-2019, NVIDIA Corporation. All rights reserved.