Dear all,
the scheduled maintenance has been completed and the cluster is back in full operation. Below is a summary of the changes introduced.
Action required - Default Python module changed
The anaconda3 module is no longer loaded by default when you open a shell. It has been replaced by python/3.11.11-gcc-11.4.0, managed via Spack.
If your workflows relied on Anaconda being available by default, you have three options:
- Load it explicitly in your scripts or session:
module load anaconda3 - Save it to your personal default module collection:
module load anaconda3 && module save - Add
module load anaconda3to your~/.bashrcfor automatic loading at login
CUDA is still loaded by default — no action needed on that front.
New compute nodes
Three new worker nodes have been added to the cluster: mereu, dracula, and bimbogigi, funded through the FAIR PNRR project. Each node is equipped with 80 CPUs, ~512 GB RAM, and 4× NVIDIA L40S 45 GB GPUs. The boost_usr_prod partition is now significantly larger, which should noticeably reduce queuing times for high-end GPU jobs.
New /dres filesystem
A new storage pool, /dres, is now mounted on the login nodes (ailb-login-02, ailb-login-03). Unlike /work, which is tied to the project lifecycle, /dres is intended for long-term retention of final project results. When a project concludes, the PI can request migration of data from /work to /dres for permanent archival. Storage in /dres is not deleted on project expiry.
If you are interested in long-term storage for your project, contact the HPC Helpdesk.
Package upgrades and documentation
All system packages have been upgraded to their latest versions. Over the past few days, the user documentation has also been greatly expanded - new articles cover topics such as job submission, storage management, Python environments, containerisation, and more. We encourage you to explore it; you may find answers to questions you didn’t know you had.
For any questions or issues, do not hesitate to contact us.
Best regards,
The AImageLab-HPC Team