ChangeLog¶

Updates to ScienceCloud and ScienceCluster are available on this page. For the Supercomputer service, please see the Alps User Guide.

2025¶

December¶

2025-12-04¶

ScienceCluster:
- ScienceCluster is migrated to the new ScienceCloud with upgrades to operating system, Slurm, system packages and drivers.
- If you receive the Remote host identification has changed error when connecting with ssh, please run ssh-keygen -R cluster.s3it.uzh.ch, and see here for more info.
- Hardware changes:
  - CPU: Upgrade to AMD Epyc 9004 processors with higher-speed RAM
  - GPU: Addition of L4, H100 and H200 compute nodes. Removal of T4 nodes.
- Storage changes: /home/$USER and /data/$USER are merged into single workspace. /data/$USER is now /home/$USER/data. Combined quota on /home is 400 GB, with no limit on number of files.
- Software changes:
  - Miniforge replaces Anaconda, see our guide how to use Miniforge.
  - Apptainer replaces Singularity, see our guide for how to use Apptainer.
  - Use module load dev2025a make available additional modules such as openmpi, cmake, fftw, hdf5, etc.
  - (ScienceApps) includes Remote Desktop Environments (Xfce and MATE) with GPU computing support, see our guide for documentation on how to use Xfce and MATE.
- Modules:
  - New released modules: miniforge3 1.4.1, apptainer 25.3.0-3, jupyter 2025-11-25, rstudio 4.5.2, VS Code 1.106.2 (code-server 4.106.2), matlab 2024b, tensorboard 2.20.0, mate 24.04-2025b, xfce 24.04-2025b.
  - Removed modules: python 3.13.0 (and all earlier versions, use uv), anaconda3 2024.02-1 (and all earlier versions), mamba 24.9.0-0 (and all earlier versions), singularityce 4.2.1 (and all earlier versions).
  - If a command such as module available gives you an error, please add the flag to ignore the cache: module --ignore_cache available and perhaps remove the cache content in either ~/.lmod/ or ~/.lmod.d/

April¶

2025-04-02¶

ScienceCluster:
- Hardware changes: GPU H100 nodes added.

March¶

2025-03-05¶

ScienceCluster:
- Quota on /shares: to ensure system stability, quotas have been implemented on all top-level project folders. These quotas are set to 10 TiB above the current usage. The quota is hard, meaning all writing activity will be blocked immediately once the system detects an overage. You can request a quota increase for your /shares/<project name> folder at any time by contacting Science IT. All requests will be granted as long as capacity allows.

January¶

2025-01-16¶

ScienceCluster:
- scrontab has been enabled for scheduling Slurm jobs, and regular crontab for users has been deactivated. See our guide for documentation on how to use scrontab.

2024¶

November¶

2024-11-06¶

ScienceCluster:
- Users can get the updated list of modules and avoid module load errors by removing the lmod cache. We therefore suggest to run the following command:
```
rm .cache/lmod/*
```
- New released modules: VS Code 1.93 (code-server 4.93.1), jupyter 2024-10-22, RStudio Server with R 4.4.2, mamba 24.9.0-0, rclone 1.68.1, singularityce 4.2.1, boost 1.86.0, perl 5.40.0, cmake 3.30.5, gsl 2.8, python 3.13.0, nvhpc 24.9, cuda 12.6.2, nccl 2.22.3-1, cudnn 9.5.1.17-12, gcc 14.2.0, openmpi 5.0.5, hdf5 1.14.5, intel-oneapi-compilers 2025.0.0, intel-oneapi-mkl 2025.0.0, intel-oneapi-tbb 2022.0.0, intel-oneapi-mpi 2021.14
- Removed unused and old modules: Matlab 2021b, VS Code Server 3.10.2, anaconda3 2022.05, 2022.10, mamba 23.3.1-1, perl 5.34.1, rclone 1.59.1, singularity 3.10.2, 3.11.3, 4.0.2, 4.1.0, cmake 3.26.3
- Upgrade: NVIDIA driver

August¶

2024-08-22¶

ScienceCluster:
- (ScienceApps) Upgraded VSCode. Added support for interactive GPU sessions.

July¶

2024-07-03¶

ScienceCluster:
- Upgraded Slurm to version 24.05
- New released modules: cmake 3.29.6, cudnn 9.2.0.82-12 (with cuda 12.4.1), gcc 14.1.0 (with cuda 12.4.1), openmpi 5.0.3, fftw 3.3.10 and hdf5 1.14.3, both recompiled with openmpi 5.0.3

June¶

2024-06-05¶

ScienceCluster:
- The default value of 1MB is now set for the Slurm memory parameters. This will prevent the whole node allocation when no memory parameters is specified while running sbatch/srun commands.
  
  Important: adjust your Slurm submission scripts so that one of the memory parameters is specified (--mem, --mem-per-cpu, --mem-per-gpu), otherwise the default value of 1MB will be used.
- New released modules: python 3.12.3, anaconda3 2024.02-1, mamba 24.3.0-0, cuda 12.4.1, cudnn 8.9.7.29-12, jupyter 2024.06 (4.2.1), RStudio Server 2024.06 (R 4.4)
- Removed unused and old modules: boost 1.78.0, boost 1.83.0, python 3.11.0, mamba 4.14.0-0, mamba 22.9.0-2, mamba 23.1.0-1, nvhpc 22.7, rclone 1.62.2
- Hardware changes: replacement of the large memory nodes and several CPU nodes of medium memory added.

May¶

2024-05-08¶

ScienceCluster:
- System upgrades for improved stability and security
- singularityce 4.1.0 is now the default version

March¶

2024-03-06¶

ScienceCloud:
- Access to the the cloud console (cloud.s3it.uzh.ch) has been restricted to UZH-only networks (wired, wifi, and VPN). The cloud console is now dropping external non-UZH connection attempts.

February¶

2024-02-07¶

ScienceCluster:
- Users can now use the command "ls -lh" to print the content size for each folder/file: this allows to quickly find large folders in case of over-quota issues, including hidden ones (e.g. ".local").
- Upgrade: NVIDIA driver
- New released modules: cuda 12.3.0, openmpi 5.0.1, boost 1.84.0, mamba 23.11.0-0, rclone 1.65.1, singularityce 4.1.0

2023¶

December¶

2023-12-04¶

ScienceCluster:
- Updates to modules: python 3.12, anaconda3 2023.09-0, mamba 23.3.1-1, nvhpc 23.9, singularityce 4.0.2, cuda 12.2.1, nccl 2.19.3-1, cudnn 8.9.5.30-12.2, cudnn 8.9.5.30-11.8, gcc 13.2.0, openmpi 4.1.6

2023-11-09¶

ScienceCluster:
- Matlab R2023b module released
ScienceCloud:
- Images updated; see here for the list of active images
  - MatLab and Gaussian users should use a "***Singularity..." image then follow these MatLab and/or Gaussian instructions

2023-11-01¶

ScienceCluster:
- OnDemand (for ScienceApps) upgraded for improved stability and security
- Upgraded kernel version and firmware on compute nodes for improved stability and security
- Users can now take advantage of all GPU nodes via the lowprio partition