Our central clients are powerful computers in the server room, running 24-7. The software available is the same as on the Fedora Linux desktops and laptops at the D-MATH. We distinguish between two types of central clients: compute clients and load clients.
Compute clients
Note: The central IT services have a very powerful HPC-cluster called Euler.
Application area: compute jobs like R, Python or Matlab which produce heavy load or use a lot of memory.
Important: Run your compute jobs with scheduling priority 7.
Instead of : $ <command-line>
use on bash : $ nice -n 7 <command-line>
Therefore, utility programs like an editor (emacs, vi) are working even on a compute client with a high load.
| Name | CPU / GPU | RAM | Users |
|---|---|---|---|
| ada-20 | 2x Intel Xeon E5-2699 v4 (22 Cores) 2.20GHz | 512 GB | SfS only |
| ada-21 | 2x Intel Xeon E5-2699 v4 (22 Cores) 2.20GHz | 512 GB | all |
| ada-22 | 2x Intel Xeon E5-2697 v4 (18 Cores) 2.3 GHz 1x nVidia GeForce GTX 1080 Ti (12GB) 1x nVidia GeForce GTX 1080 (8GB) | 256 GB | all |
| ada-23 | 2x Intel Xeon Gold 6148 (20 Cores) 2.4 GHz 4x nVidia GeForce GTX 1080 Ti (12 GB) | 256 GB | all |
| ada-24 | 2x Intel Xeon Gold 6252 (24 Cores) 2.1 GHz 2x nVidia GeForce RTX 2080 Ti (12 GB) | 512 GB | all |
| ada-25 | 2x Intel Xeon Gold 6254 (18 Cores) 3.1 GHz 2x nVidia GeForce RTX 2080 Ti (12 GB) | 512 GB | all |
| ada-26 | 1x AMD EPYC 7742 (64 Cores) 2.25GHz 2x nVidia GeForce RTX 3090 (24 GB) | 512 GB | all |
| ada-27 | 1x AMD EPYC 7763 (64 Cores) 2.45GHz 2x nVidia GeForce RTX 4090 (24 GB) | 512 GB | all |
| ada-28 | 2x AMD EPYC 7313 (16 Cores) 3.0GHz 4x nVidia GeForce RTX 4090 (24 GB) | 1'024 GB | SfS only |
| ada-29 | 1x AMD EPYC 7763 (64 Cores) 2.45GHz 2x nVidia RTX 6000 Ada (48 GB) | 512 GB | all |
| ada-30 | 1x AMD EPYC 9554 (64 Cores) 3.1GHz 3x nVidia L40s (48GB) | 1'152 GB | SfS only |
| ada-31 | 1x AMD EPYC 9554 (64 Cores) 3.1GHz 2 x nVidia RTX PRO 6000 Blackwell (96GB) | 1'152 GB | all |
| ada-32 | 1x AMD EPYC 9554 (64 Cores) 3.1GHz 2 x nVidia RTX PRO 6000 Blackwell (96GB) | 1'152 GB | all |
Current usage
You check the current usage of all our compute clients with the command ada-info, which is available on all out Linux systems. To run it from a macOS, creare for instance an alias like this in your ~/.bashrc or ~/.zsh:
alias ada-info='ssh ion-1 ada-info'
The output looks like this:
| host | thr | gpu | free/tot mem | swap | load average |
| ada-20 | 88 | 0 | 348/503 GB | 0.0 | 0.10, 0.07, 0.08 |
| ada-21 | 88 | 0 | 14/503 GB | 23.6 | 2.14, 2.16, 2.12 |
| ada-22 | 72 | 2 | 234/251 GB | 0.0 | 0.00, 0.00, 0.00 |
| ada-23 | 80 | 4 | 242/251 GB | 0.0 | 0.00, 0.00, 0.00 |
...
See ada-info -h for more options.
Load clients
Application area: Programs you need for your desktop environment but use too much power on your computer, like firefox, libreoffice or texing large documents. Do not use these clients for your compute jobs!
For example:
ssh ion-1
pdflatex my-phd.tex
| Name | Type | CPU | RAM | Users |
|---|---|---|---|---|
| ion-1 | Virtual | 4 cores | 12 GB | all |
| ion-2 | Virtual | 4 cores | 12 GB | all |
| ion-3 | Virtual | 4 cores | 12 GB | all |
| ion-4 | Virtual | 4 cores | 12 GB | all |
Local data
If you want to save data locally (which is faster than on the network), you can do this in /userdata. E.g.:
ssh ada-31
cd /userdata/$USER
# do your stuff here
To access the local data of an ada-XX from another ada (or from a linux desktop computer), go to /r/ada-XX/$USER. Of course accessing the data in this way is slower. E.g.:
ssh ada-31
cd /r/ada-31/$USER
# you now see your stuff of ada-31
