hpcwiki

Benutzer-Werkzeuge

Webseiten-Werkzeuge

Zuletzt angesehen:
read_hpc_blender_short

Unterschiede

Hier werden die Unterschiede zwischen zwei Versionen angezeigt.

Link zu dieser Vergleichsansicht

Nächste Überarbeitung 		Vorhergehende Überarbeitung
read_hpc_blender_short [2021/12/02 18:07]
schwan angelegt 		read_hpc_blender_short [2021/12/02 18:11] (aktuell)
schwan
Zeile 3: Zeile 3:
  
 ## Basics ## Basics
-There are two possibilities to access the system, either via SSH with username@cslogin.tu-ilmenau.de or via an interactive desktop session to be started via the browser interface (just click on "Desktop Session" to start a new one).+There are two possibilities to access the system, either via SSH with username@cslogin.tu-ilmenau.de or via an interactive desktop session to be started via the browser interface (just click on "Desktop Session" to start a new one).\\
 You need to be inside the university network if you want to access the system, so at home use the VPN. You need to be inside the university network if you want to access the system, so at home use the VPN.
-If you want to use the interactive Desktop Session, it is strongly recomended to install the DCV client and configure the "Viewer Type" of interactive desktop sessions via "Settings" to "Desktop Client" in NICE (see https://www1.tu-ilmenau.de/hpcwiki/doku.php?id=rem_vis).+If you want to use the interactive Desktop Session, it is strongly recomended to install the DCV client and configure the "Viewer Type" of interactive desktop sessions via "Settings" to "Desktop Client" in NICE (see [[rem_vis|Remote Visualization]]).
  
-It is very important to know that there are two different "types" of computing nodes: interactive nodes and non-interactive nodes. +It is very important to know that there are two different "types" of computing nodes: interactive nodes and non-interactive nodes. Interactive nodes can be accessed via SSH, to run something on non-interactive nodes you would need to use the batch system. Usually you only have access to the internet from **interactive nodes**.
-Interactive nodes can be accessed via SSH, to run something on non-interactive nodes you would need to use the batch system. +
-Usually you only have access to the internet in **interactive nodes**.+
 That means, you need to configure everything in the interactive nodes and then run it either via the batch system or on a interactive GPU node. That means, you need to configure everything in the interactive nodes and then run it either via the batch system or on a interactive GPU node.
  
Zeile 26: Zeile 24:
 If your software uses the pytorch package, it is strongly recommended to install one of the most recent pytorch versions with the pip command you can get from the official website https://pytorch.org/get-started/locally If your software uses the pytorch package, it is strongly recommended to install one of the most recent pytorch versions with the pip command you can get from the official website https://pytorch.org/get-started/locally
  
-A valid command to install pytorch would be e.g. `pip3 install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html`+A valid command to install pytorch would be e.g. **pip3 install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html**\\
  
-Take care to run your Python package installations on an interactive node, if you do it e.g. on makalu86 it won't work because you don't have internet access on this node.+ 
 +Take care to run your Python package installations on an interactive node, if you do it e.g. on ??? it won't work because you don't have internet access on this node.
 Even if your software uses an older pytorch version, it is usually worth it to try out a recent version supporting CUDA version >= 11.0. Even if your software uses an older pytorch version, it is usually worth it to try out a recent version supporting CUDA version >= 11.0.
 It is important to have a pytorch version with CUDA >= 11.0 as the NVIDIA A100 graphics cards only support CUDA >= 11.0. It is important to have a pytorch version with CUDA >= 11.0 as the NVIDIA A100 graphics cards only support CUDA >= 11.0.
Zeile 34: Zeile 33:
 ## Interactive GPU node ## Interactive GPU node
 The easiest way to get started with running some GPU-based computing stuff is to access the interactive GPU node. The easiest way to get started with running some GPU-based computing stuff is to access the interactive GPU node.
-There is currently one interactive GPU node with 4x NVIDIA A100 graphics cards, the name of this machine is makalu86. +There is currently one interactive GPU node with 4x NVIDIA A100 graphics cards, the name of this machine is ...
-**Important:** due to security reasons, **interactive GPU nodes don't have access to the internet** so you need to download/prepare everything on a standard interactive node and then execute your command "offline" on the interactive GPU node. +
-Further, you can access the interactive GPU nodes only if you already are logged in into an interactive node (either via SSH or via an interactive Session).+
  
-To access an interactive GPU node, enter `ssh makalu86`, type in your CS login password and there you are.+ 
 +**Important:** due to security reasons, **interactive GPU nodes don't have access to the internet** so you need to download/prepare everything on a standard interactive node and then execute your command "offline" on the interactive GPU node. Further, you can access the interactive GPU nodes only if you already are logged in into an interactive node (either via SSH or via an interactive Session). 
 + 
 +To access an interactive GPU node, enter ???, type in your CS login password and there you are.
 To check if you are on the correct node, enter `nvidia-smi`, four A100 graphics cards should appear. To check if you are on the correct node, enter `nvidia-smi`, four A100 graphics cards should appear.
 Now you can start to test your scripts, etc. Now you can start to test your scripts, etc.
Zeile 45: Zeile 45:
  
 ### Estimate efficiency of GPU usage ### Estimate efficiency of GPU usage
-On makalu86, `nvidia-smi` is for sure the best way to monitor how good your software is using the GPU.+On ???, `nvidia-smi` is for sure the best way to monitor how good your software is using the GPU.
 Here, the power usage is for sure the best way to find out how efficient your software is using the GPU. Here, the power usage is for sure the best way to find out how efficient your software is using the GPU.
 Another good indicator is the temperature, if the GPU is not busy it will have around 33°, starting from ~60° you can say the GPU is getting busy. Another good indicator is the temperature, if the GPU is not busy it will have around 33°, starting from ~60° you can say the GPU is getting busy.
 +
  
 ## GPU Batch system ## GPU Batch system
read_hpc_blender_short.1638464843.txt.gz · Zuletzt geändert: 2021/12/02 18:07 von schwan

Seiten-Werkzeuge

Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki