AMS

The Amsterdam Modeling Suite (AMS) is a comprehensive suite of computational chemistry software. It comes with a graphical user interface, AMS-GUI that provides an intuitive way for setting up, running, and analyzing simulations.

Available

• Puhti: AMS, Version 2024.102, 2025.105
• Mahti: AMS, Version 2024.102, 2025.105
• Roihu-CPU: AMS, Version 2026.104

License

• The license entitles software usage by any academic researcher or student of an academic institute where "Academic" means "belonging to a degree-granting institute".
• The license does not include the right for employees of government labs or other non-academic non-profit research institutes to use the software.
• The license only allows non-profit non-commercial use.
• The license excludes all forms of contract research, royalty-bearing activities and other activities leading to monetary benefits.
• Research groups which need other modules of the AMS suite can obtain a license for themselves to run the programs on the CSC computers.

Usage

Initialize AMS on Puhti and Mahti:

module load ams/2025.105

Initialize AMS on Roihu-CPU:

module load ams/2026.104

Example batch scripts

Note

Particularly some property calculations can be very disk I/O intensive. Such jobs benefit from using the fast local storage (NVMe) on Puhti, Mahti, or Roihu. Using local disk for such jobs will also reduce the load on the Lustre parallel file system.

PuhtiPuhti, local diskMahtiMahti, local diskRoihu-CPURoihu-CPU, local diskRoihu-CPU, disaggregated NVMe

#!/bin/bash
#SBATCH --partition=test
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=40      # MPI tasks per node
#SBATCH --account=yourproject     # insert here the project to be billed 
#SBATCH --time=00:15:00           # time as `hh:mm:ss`
#SBATCH --mem-per-cpu=1500        # requested memory per process in MB
module purge
module load ams/2025.105
export SCM_TMPDIR=$PWD/$SLURM_JOB_ID
mkdir -p $SCM_TMPDIR
# Create an example input file from the examples 
sed '1,4d;$d;/Print/,/End/d' $AMSHOME/examples/Benchmarks/ADF/Si35_TZ2P/Si35_TZ2P.run  > ./Si35_TZ2P.inp
"$AMSBIN/ams" < ./Si35_TZ2P.inp > ./Si35_TZ2P.log

#!/bin/bash
#SBATCH --partition=large
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=40      # MPI tasks per node
#SBATCH --account=yourproject     # insert here the project to be billed
#SBATCH --time=00:15:00           # time as `hh:mm:ss`
#SBATCH --mem-per-cpu=1500        # requested memory per process in MB
#SBATCH --gres=nvme:100           # requested local disk space in GB
module load ams/2025.105
export SCM_TMPDIR=$LOCAL_SCRATCH
# Create an example input file from the examples
sed '1,4d;$d;/Print/,/End/d' $AMSHOME/examples/Benchmarks/ADF/Si35_TZ2P/Si35_TZ2P.run  > ./Si35_TZ2P.inp
"$AMSBIN/ams" < ./Si35_TZ2P.inp > ./Si35_TZ2P.log

#!/bin/bash
#SBATCH --partition=medium
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=128 # MPI tasks per node
#SBATCH --account=yourproject # insert here the project to be billed
#SBATCH --time=00:20:00       # time as `hh:mm:ss`
module purge
module load ams/2025.105
export SCM_TMPDIR=$PWD/$SLURM_JOB_ID
mkdir -p $SCM_TMPDIR

# Create an example input file from the examples
sed '1,4d;$d;/Print/,/End/d' $AMSHOME/examples/Benchmarks/ADF/Si35_TZ2P/Si35_TZ2P.run  > ./Si35_TZ2P.inp
"$AMSBIN/ams" < ./Si35_TZ2P.inp > ./Si35_TZ2P.log

#!/bin/bash
#SBATCH --partition=small
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=128 # MPI tasks per node
#SBATCH --account=yourproject # insert here the project to be billed
#SBATCH --time=00:20:00       # time as `hh:mm:ss`
#SBATCH --gres=nvme:100
module purge
module load ams/2025.105
export SCM_TMPDIR=$LOCAL_SCRATCH
mkdir -p $SCM_TMPDIR

# Create an example input file from the examples
sed '1,4d;$d;/Print/,/End/d' $AMSHOME/examples/Benchmarks/ADF/Si35_TZ2P/Si35_TZ2P.run  > ./Si35_TZ2P.inp
"$AMSBIN/ams" < ./Si35_TZ2P.inp > ./Si35_TZ2P.log

#!/bin/bash
#SBATCH --partition=small
#SBATCH --nodes=1
#SBATCH --ntasks=128              # MPI tasks
#SBATCH --account=yourproject     # insert here the project to be billed
#SBATCH --time=00:20:00           # time as `hh:mm:ss`
#SBATCH --mem-per-cpu=1500M       # requested memory per process
module purge
module load ams/2026.104
export SCM_TMPDIR=$PWD/$SLURM_JOB_ID
mkdir -p $SCM_TMPDIR
# Create an example input file from the examples
sed '1,4d;$d;/Print/,/End/d' $AMSHOME/examples/Benchmarks/ADF/Si35_TZ2P/Si35_TZ2P.run > ./Si35_TZ2P.inp
"$AMSBIN/ams" < ./Si35_TZ2P.inp > ./Si35_TZ2P.log

Choosing core counts on Roihu-CPU

Roihu M-nodes have 384 cores, but ADF does not scale efficiently to a full node for typical jobs. The table below shows scaling for the Si35_TZ2P benchmark — note that scaling is strongly system- and model-dependent, so these numbers should be used as a guide rather than absolute values. 128–192 cores gives the best balance of efficiency and speed for this case. Beyond that, wall time barely improves while memory usage and billing costs increase significantly. For full-node jobs, use local NVMe scratch (see the "Roihu-CPU, local disk" tabs) to avoid I/O contention from AMS's per-rank scratch files.

MPI ranks	CPU efficiency	Wall time
48	~95%	~437 s
96	~91%	~385 s
128	~90%	~302 s
192	~85%	~252 s
384 (full node)	~75%	~247 s

#!/bin/bash
#SBATCH --partition=small
#SBATCH --nodes=1
#SBATCH --ntasks=128              # MPI tasks
#SBATCH --account=yourproject     # insert here the project to be billed
#SBATCH --time=00:20:00           # time as `hh:mm:ss`
#SBATCH --mem-per-cpu=1500M       # requested memory per process
module purge
module load ams/2026.104
# On Roihu, local NVMe scratch is available via $TMPDIR (no --gres flag needed)
export SCM_TMPDIR=$TMPDIR/ams_$SLURM_JOB_ID
mkdir -p $SCM_TMPDIR
# Create an example input file from the examples
sed '1,4d;$d;/Print/,/End/d' $AMSHOME/examples/Benchmarks/ADF/Si35_TZ2P/Si35_TZ2P.run > ./Si35_TZ2P.inp
"$AMSBIN/ams" < ./Si35_TZ2P.inp > ./Si35_TZ2P.log

See also

See the "Roihu-CPU" tab for core-count recommendations. For full-node jobs requiring maximum I/O performance, consider the disaggregated NVMe option in the "Roihu-CPU, disaggregated NVMe" tab.

Maximum amount of local NVMe disk available in $TMPDIR depends on the Slurm partition and node type. On shared nodes (e.g. on the small and longrun partitions), you have access to 20 GiB of tmp storage. On full nodes (medium and large partitions), you have access to 600 GiB of local storage. See Roihu storage documentation for more details.

Disaggregated NVMe on Roihu

Roihu provides a centralised pool of 307.2 TB fast NVMe storage served over InfiniBand NDR. For I/O-intensive full-node jobs this can outperform both Lustre scratch and node-local $TMPDIR. It requires a full-node partition (medium or large) and a #BB_LUA burst-buffer directive — shared-node support is not yet available. See Roihu disk areas for full details.

#!/bin/bash
#SBATCH --partition=medium
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=384     # full node required for disaggregated NVMe
#SBATCH --account=yourproject     # insert here the project to be billed
#SBATCH --time=00:20:00           # time as `hh:mm:ss`
#BB_LUA SBF storagesize=100GB path=/run/sbb/<user>
module purge
module load ams/2026.104
# Disaggregated NVMe is mounted at the path specified in #BB_LUA above
export SCM_TMPDIR=/run/sbb/<user>/ams_$SLURM_JOB_ID
mkdir -p $SCM_TMPDIR
# Create an example input file from the examples
sed '1,4d;$d;/Print/,/End/d' $AMSHOME/examples/Benchmarks/ADF/Si35_TZ2P/Si35_TZ2P.run > ./Si35_TZ2P.inp
"$AMSBIN/ams" < ./Si35_TZ2P.inp > ./Si35_TZ2P.log

Benchmark results for Si35_TZ2P at 384 ranks

Benchmarking showed that disaggregated NVMe gives the best performance for full-node ADF jobs, outperforming both Lustre scratch and node-local $TMPDIR:

Scratch location	Wall time	CPU efficiency	Billing units
Lustre scratch	~247 s	~75%	~20 BU
$TMPDIR (node-local)	~230 s	~91%	~19 BU
Disaggregated NVMe	~190 s	~89%	~15 BU

Note that scaling is strongly system- and model-dependent. See the "Roihu-CPU" tab for guidance on choosing core counts.

The AMS-GUI

AMS comes with an integrated Graphical User Interface, AMS-GUI that makes it easy to set up, run and analyze modelling tasks. You can test the GUI via the Puhti web interface, but for more extensive use we recommend installing the GUI on your own laptop/workstation. For detailed instructions, see the AMS-GUI documentation.

References

Depending on your usage, be careful to properly cite the AMS driver, used calculation engines as well as feature references. For details, see the relevant AMS documentation

More information

• AMS support pages
• Tutorials
• Workshops
• FAQ
• Knowledge bank
• The latest news

Do you find this page useful?

Yes

Thank you for your feedback!

No

Thank you! Your feedback is appreciated.

Please let us know how we could improve this page by contacting us or by creating an issue in GitHub. You can also edit this page yourself and contribute your improvements by creating a pull request.