Skip to content

C I R R U S

Research and Development Node

Access
  1. Request access / will be done for you by your supervisor.
  2. As staff, access using SSH - How to SSH / VNC / VPN
  3. Inform yourself: Getting Started

Welcome to the HPC @IMG @UNIVIE and please follow these steps to become a productive member of our department and make good use of the computer resources. Efficiency is keen.

System information

Name Value
Product G294-Z42-AAP2-000
CPU model AMD EPYC 9355 32-Core Processor
Cores 2 CPU, 32 physical cores per CPU, total 64 logical CPU units
Memory 24x 64GB - 1536 GB Total
Memory/Core 23 GB
OS RockyLinux 10.1 Red Quartz
Purchase December 2025
OOB Management
IP 131.130.157.9
C-Name cirrus.img.univie.ac.at
Cirrus
1
2
3
4
5
6
7
8
             o\ cirrus
   _________/__\__________
  |  -- ( -~- ) --   - (  |
  |    ~-  -  ~-      . `-|
  |                  '   ||
  |                       |
  |  131.130.157.9        |
  |_______________________|

The server is connected to the SRV Storage System via AURORA by infiniband FDR-40.

Software

The server is intended as a GPU machine and has CUDA installed as well as the nvidia-compilers and gnu compilers.

Major librires:

  • OpenMPI (5.0.9)
  • HDF
  • NetCDF (C, Fortran)
  • ECCODES from ECMWF
  • Math libraries e.g. intel-mkl, lapack, ...
  • Interpreters: Python
  • Tools: cdo, ncl, nco

These software libraries are usually handled by environment modules.

Currently installed modules

Please note that new versions might already be installed.

available modules
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
$ module av
- /home/swd/spack/share/spack/modules/linux-rocky10-x86_64 -
gcc/14.3.1-ioptbfu  

-- /home/swd/spack/share/spack/modules/linux-rocky10-zen5 --
automake/1.18.1-gcc-14.3.1-bqwspo5            
cdo/2.5.4-gcc-14.3.1-7h3j7w4                  
cuda/12.9.1-none-none-yob6yrd                 
cuda/13.0.2-none-none-rbdlhdl                 
eccodes/2.45.0-gcc-14.3.1-t2h67kn             
fftw/3.3.10-gcc-14.3.1-ir2r4aw                
fftw/3.3.10-gcc-14.3.1-tamvrit                
gcc/15.2.0-dtlrmiv                            
geos/3.14.1-gcc-14.3.1-c2yyfca                
hdf5/1.14.6-gcc-14.3.1-lt746n3                
intel-oneapi-mkl/2024.2.2-gcc-14.3.1-smhtvuo  
m4/1.4.20-gcc-14.3.1-rqxc7h4                  
miniconda3/25.5.1-none-none-ihs5iwb           
nco/5.3.4-gcc-14.3.1-g3wpoha                  
ncview/2.1.9-gcc-14.3.1-4ubp57u               
netcdf-c/4.9.2-gcc-14.3.1-rqhorll             
netcdf-fortran/4.6.2-gcc-14.3.1-5k465xw       
netlib-lapack/3.12.1-gcc-14.3.1-q5afu6p       
netlib-scalapack/2.2.2-gcc-14.3.1-x3j2dg4     
nvhpc/25.11-none-none-k6f25rn                 
openblas/0.3.30-gcc-14.3.1-m3mn4mr            
openmpi/5.0.9-gcc-14.3.1-kbek334              
parallel-netcdf/1.14.1-gcc-14.3.1-67ipmv7     
proj/9.7.0-gcc-14.3.1-7kznn7o                 
pypy/3.10-v7.3.19-gcc-14.3.1-yxc4cea          
python/3.14.2-gcc-14.3.1-gh4ya5v              

-------------------- /home/swd/modules ---------------------
gcc-stack/14.3.1  micromamba/2.5.0

GPU / CUDA

Currently, there are two CUDA versions installed, but since GPU applications depend a lot on the exact version of the library, please report any missing versions.

The GPU installed is a NVIDIA RTX PRO 6000 Blackwell Server Edition. with 96GB of GDDR7 memory.

The current configuration splits this rather large GPU into 3 logical parts, which can be undone at any time in the future to allow one user to take it all:

  • MIG 2g.48gb Device 0: (UUID: MIG-331f1801-304d-5ca7-934f-4e9ffda7f06a)
  • MIG 1g.24gb Device 1: (UUID: MIG-c0f972a7-9639-55fb-81f5-3741fa7b92a0)
  • MIG 1g.24gb Device 2: (UUID: MIG-715e42be-6ee8-5460-88ae-fb2d567cc82f)

So there is one larger part (48GB) and two samller parts (24GB), which are so called MIG (Multi Instance GPU) and each MIG can mostly be used by one application. It might be possible to load two or mor applications into one mig, but make sure that these do not interfere with the memory available to those. As on and off loading to GPU memory is a performance penalty, which might increase your overall runtime. For HPC applications, this is different, as one use would reserve a whole GPU or dozens.

To get started run:

userservices gpu
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
[user@cirrus]# userservices gpu
[USERSERVICE] gpu
GPU 0: NVIDIA RTX PRO 6000 Blackwell Server Edition (UUID: GPU-7d3e3a53-6844-27b1-0690-856fe0deab22)
  MIG 2g.48gb     Device  0: (UUID: MIG-331f1801-304d-5ca7-934f-4e9ffda7f06a)
  MIG 1g.24gb     Device  1: (UUID: MIG-c0f972a7-9639-55fb-81f5-3741fa7b92a0)
  MIG 1g.24gb     Device  2: (UUID: MIG-715e42be-6ee8-5460-88ae-fb2d567cc82f)
################################################################################
User Services   -  GPU overview

  userservices gpu -l

Options:
    -l                 List available GPUs
    -v                 Show GPU driver version

Examples:
    # List available GPUs (here 1GPU, but split into 3 instances for parallel use)
    userservices gpu -l  or nvidia-smi -L
    GPU 0: NVIDIA RTX PRO 6000 Blackwell Server Edition (UUID: GPU-7d3e3a53-6844-27b1-0690-856fe0deab22)
        MIG 2g.48gb     Device  0: (UUID: MIG-331f1801-304d-5ca7-934f-4e9ffda7f06a)
        MIG 1g.24gb     Device  1: (UUID: MIG-c0f972a7-9639-55fb-81f5-3741fa7b92a0)
        MIG 1g.24gb     Device  2: (UUID: MIG-715e42be-6ee8-5460-88ae-fb2d567cc82f)

    # Show what is running on the GPUs
    nvidia-smi

Usage Example:
    # How to use the GPUs?
    # This makes use of one of the MIG - GPU instances.
    CUDA_VISIBLE_DEVICES=MIG-331f1801-304d-5ca7-934f-4e9ffda7f06a python my_gpu_script.py
    # monitor with nvidia-smi -l

################################################################################
Author: MB
Date: 14.01.2026
Contact: it.img-wien@univie.ac.at
Path: /home/swd/userservices/userservices.d
################################################################################

This should give you an idea on how to run an application with the GPU. An example is given here:

gpu example
1
2
# This selects one MIG for the application
[user@cirrus]# CUDA_VISIBLE_DEVICES=MIG-331f1801-304d-5ca7-934f-4e9ffda7f06a python my_gpu_script.py

Check if nothing is running on the GPU?

gpu monitoring
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
[user@cirrus]# nvidia-smi
Mon Mar  2 14:16:01 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 590.48.01              Driver Version: 590.48.01      CUDA Version: 13.1     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA RTX PRO 6000 Blac...    On  |   00000000:18:00.0 Off |                   On |
| N/A   34C    P0             89W /  600W |   11078MiB /  97887MiB |     N/A      Default |
|                                         |                        |              Enabled |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| MIG devices:                                                                            |
+------------------+----------------------------------+-----------+-----------------------+
| GPU  GI  CI  MIG |              Shared Memory-Usage |        Vol|        Shared         |
|      ID  ID  Dev |                Shared BAR1-Usage | SM     Unc| CE ENC  DEC  OFA  JPG |
|                  |                                  |        ECC|                       |
|==================+==================================+===========+=======================|
|  0    1   0   0  |           10962MiB / 48512MiB    | 94      0 |  2   2    2    0    2 |
|                  |               0MiB / 16653MiB    |           |                       |
+------------------+----------------------------------+-----------+-----------------------+
|  0    5   0   1  |              58MiB / 24192MiB    | 46      0 |  1   1    1    0    1 |
|                  |               0MiB /  8326MiB    |           |                       |
+------------------+----------------------------------+-----------+-----------------------+
|  0    6   0   2  |              58MiB / 24192MiB    | 46      0 |  1   1    1    0    1 |
|                  |               0MiB /  8326MiB    |           |                       |
+------------------+----------------------------------+-----------+-----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0    1    0          1434068      C   ...are/mamba/envs/gpu/bin/python      10838MiB |
+-----------------------------------------------------------------------------------------+

# continuously monitoring until CTRL+C
[user@cirrus]# nvidia-smi -l

Currently the driver 595.45.04 is installed as well as CUDA 13.1 from the repository. This might change in the future to make sure that the GPU can reach it's full potential as the driver evolves.