Knowledge Base

AIFS Single 1.0 on RACC2

This article is contributed by Robert Lee. Please contact the author directly for any AIFS related questions and comments. For all other RACC2 or research computing related queries, as always please raise them to the Digital Research Computing Team via the Self Service Portal.

AIFS Single v1 is the first fully operational deterministic machine‑learning forecast model released by ECMWF on 25 February 2025. It replaces the earlier experimental version (v0.2.1) and is implemented using the open-source Anemoi framework. This latest version improves upper‑atmosphere and precipitation forecasts, adds outputs like 100 m wind, snowfall, radiation, and soil variables, and marks ECMWF’s first AI-driven model integrated into its regular forecasting system. More information can be found on the AIFS Single v1 website.

How to run AIFS Single 1.0 on RACC2

First, log in to the Reading Academic Computing Cluster (RACC2). If you’re unfamiliar with our cluster, please first consult the articles in our RACC2 knowledgebase category.

AIFS Single 1.0 is installed in a Conda environment on RACC2 and can only be run on a GPU via batch jobs submitted to the GPU node. Users must use the gpuscavenger partition for this purpose.

Note: The GPU node was funded by a separate project grant. As such, access through the gpuscavenger partition is only allowed during idle periods. Jobs submitted to this partition are subject to preemption – they may be terminated and re-queued if higher-priority tasks require GPU resources.

 

Step 1: Prepare Your Job Files

A working example is available at:
/software/slurm_examples/anemoi/

This includes the following:

  • submit_AIFS_GPU.sh – a SLURM batch script to submit the job
  • run_AIFS_Single_1_12hr_test.py – a Python script that runs a 12-hour forecast and saves an output figure

You can copy the example to your own working directory and modify as needed.

 

Step 2: Submit the Batch Job

Load the pre-installed environment and submit your job:

module load anaconda/2023.09-0/anemoi
sbatch submit_AIFS_GPU.sh

This will queue your job to run on the GPU via the gpuscavenger partition.

Optional: Check Job Status
To check the status of your submitted job, run the command:

squeue -u $USER

($USER will automatically use your own username)

 

Step 3: Check the Output

Once the job completes, you will find:

  • A log file, e.g. aifs_single_1_12hr_test_output.log
  • A PNG forecast figure saved under the figures/ directory

Additional Notes

 

DTS Status Page
Drop-in Sessions

ACT services access flow

News
Suggest Content…
Click here and let us know!

Related articles

Transferring files to and from Linux storage

NAG library

ssh iconExternally visible ssh server – arc-ssh

NX authorization code

Recover a file on research storage

Using R and R Studio on RACC

ssh iconAccess to ACT Servers via an SSH Client

Running Matlab Scripts as Batch Jobs

Cronjobs on the Academic Computing Cluster

Python on the Academic Computing Cluster