-
nsys: Nvidia GPU and CPU profiler
nsys: Nvidia GPU and CPU profiler
Available
- Puhti: version depends on the CUDA environment
- Mahti: version depends on the CUDA environment
- Roihu-GPU: version depends on the CUDA environment
License
Usage is possible for both academic and commercial purposes.
Usage
NVIDIA Nsight Systems is a performance analysis tool for tracing GPU and CPU workloads. It enables the collection of a timeline of CUDA-related activities, including kernel execution, memory transfers, memory set and CUDA API calls and events or metrics for CUDA kernels. The tool is very useful in identifying the high-level bottlenecks, hotspots and for determining which kernels should be targeted for optimization and analysis with the Nsight Compute tool.
Profiling is done in two steps:
- Collect profiling data by running the application under
nsyscommand line tool - Analyze the results with
nsys-uiGUI.
To use nsys, one needs to first load the CUDA module:
For collecting data, run your application normally via Slurm and prepend your executable with nsys <profiling options>:
--argos=no needs to be provided for srun:
nsys supports many useful options. For more details please check the NVIDIA documentation.
By default, results are stored in a file named report<N>.nsys-rep, which can then be analyzed with the nsys-ui GUI:
The nsys-ui can be run directly on the CSC supercomputers, however, for smoother operation of the GUI
it is recommended to copy the results files to your local workstation and view them using a local
installation of Nsight Systems.