Programming/Nsight, TensorRT-LLM

[Nsight] Systems & Compute Install and Setting (Eng)

klue 2024. 4. 25. 03:48
반응형

updates  : 2024.04.25

contents : Install and set Nsight Systems on Docker or Linux 

# download install file from NVIDIA Nsight Systems
wget https://developer.nvidia.com/downloads/assets/tools/secure/nsight-systems/2024_1/nsightsystems-linux-public-2024.1.1.59-3380207.run
bash nsightsystems-linux-public-2024.1.1.59-3380207.run

# To uninstall the Nsight Systems 2023.4.1, please delete "/opt/nvidia/nsight-systems/2023.4.1"

# if nsys command not found,
# /root/.bashrc : add (file path is an example)
export PATH="$PATH:/opt/nvidia/nsight-systems/2024.1.1/bin"

source ~/.bashrc

nsys status --environment check needed
# if profiling error occurs, go link

 

# download install file from NVIDIA Nsight Compute
# Activate set full identifier
path : /root/Documents/NVIDIA Nsight Compute/2023.1.0/Sections/SpeedOfLight_Hierarchical*RooflineChart.section

Identifier: "SpeedOfLight_HierarchicalTensorRooflineChart"
DisplayName: "GPU Speed Of Light Hierarchical Roofline Chart (Tensor Core)"
Extends: "SpeedOfLight"
Description: "High-level overview of the utilization for compute and memory resources of the GPU presented as a roofline chart."
Order: 12
Sets {
  Identifier: "roofline"
}

#add this part
Sets {
  Identifier: "full"
}

Metrics {
 Metrics {
   Label: "Theoretical Tensor Instructions Executed"
   Name: "sm__inst_executed_pipe_tensor.sum.peak_sustained"
 }

 

반응형