Introduction¶

Summary¶

AMS2024 can be used on anything from a simple laptop to big Linux cluster environments, and is tested on a wide variety of hardware.

Recommended minimum hardware specification:

• Intel i5 or better 64bit CPU supporting AVX2 instructions

• 8GB RAM

• 250GB SSD

• OpenGL 3.2 graphics

Absolute minimum hardware specification:

• 64bit CPU supporting AVX2 instructions

• 2GB RAM

• 6GB storage for installation

• OpenGL 1.4 graphics

Supported Operating Systems:

• Windows 10/11

• macOS 10.15 or newer

• Linux with GLIBC v2.17 or higher: CentOS/RHEL 7 or newer, Debian 8 or newer, SUSE 12.3 or newer, Ubuntu 14.04 or newer, etc. AMS is regularly tested on CentOS 7, Ubuntu 20.04/22.04 and Arch Linux.

Specific hardware and software requirements for the AMS package depend on the platform. The list of supported platforms, including information on the operating system, parallel environment (MPI), and compilers used is available in the Download section of our web site.

Detailed Hardware requirements¶

CPU

AMS2024 works on modern Intel CPUs with AVX2 or AVX-512 instruction sets and AMD Zen CPUs (Ryzen, Threadripper, EPYC). Especially the AMD Zen2/Zen3 processors give very good performance (Ryzen 3000/4000/5000, EPYC 7**2/7**3). The macOS version of AMS2024 works on Apple Silicon (arm64) processors as well, but we also provide a binary for older intel (x86) macOS systems.

Memory

In a parallel calculation, the total amount of memory used by the job is a sum of that used by each process. Starting from ADF2010, some large chunks of data are placed in the shared memory so the sum rule does not strictly hold. In principle, it is more correct to count memory per process but since AMS is an MPI program it runs most efficiently when the number of processes corresponds to the number of physical processors cores. Therefore, all memory amounts below are per processor core.

The amount of RAM per core needed to run AMS depends greatly on the kind of calculation you perform. For small calculations, 256 MB will be sufficient, but if there is a lot of memory available AMS may run significantly faster. A large amount memory also reduces the load on the disks, which may speed up your calculation depending on the I/O sub-system and the operating system.

The memory requirement increases with the system size. For example, for a molecule containing 100 atoms with a DZ basis set it may be sufficient to have 256 MB but for a molecule with 1000 atoms up to 8 gigabytes may be required. Also, if you are going to perform TDDFT, relativistic spin-orbit or analytical frequency calculations then the amount of memory should be larger. As an indication, an analytical vibrational frequency calculation of a organometallic complex containing 105 atoms with a TZP basis set uses up to 1GB of RAM per process but it can be done with less, even though not as efficiently.

Disk

For installation of the package on Linux/Unix you need from about 5GB (without sources) to 8GB (with sources and compiled objects). The run-time (scratch) disk space requirements greatly depend on what type of calculation you perform. For the ADF engine, it may range from a few megabytes for a small molecule up to a hundred gigabytes for a large molecule with a large basis set, the amount scaling quadratically with the system size. Using a Solid State Drive (SSD) helps performance, especially for bigger calculations.

Network

First of all, a network card must be present in the computer as its hardware MAC address is used as the computer’s ID for the licensing.

In order to enable MPI on a standalone Windows computer, one may need to create a dummy network connection by adding a network “card” called Microsoft Loopback Adapter. This interface will be assigned an IP address from a private range.

Multi-host parallel performance.

As far as performance concerned, a switched Gigabit Ethernet network is typically sufficient for good parallel performance on up to four nodes if the nodes do not have too many CPU cores per node. If you are going to use more nodes, or nodes with high core count (>16 cores per node), you may need faster communication hardware, such as Infiniband, to get good performance. Please note that multi-host execution is not supported on Windows.

Graphics

The AMS GUI requires OpenGL 3.2 to run. For Linux and Windows users there is an OpenGL software mode available for older hardware, please read more about it in the Remote GUI documentation.

Software requirements¶

Operating System

The package runs on Windows and on the following Unix variants: Linux, Mac OS.

On the Apple systems the macOS 13.2 (Ventura) and newer is supported on Apple Silicon. For Intel-based x86-64 macOS 10.15+ (Catalina) is the minimum version.

On linux both the compute engines, python scripting environment and GUI require a GLIBC version of 2.17 or higher. AMS is compiled on CentOS 7, the code gets tested daily on CentOS 7, Ubuntu 22.04 and Arch.

The Windows version of AMS is supported on the desktop editions of Windows (10 and 11). The Windows Server is not supported.

Additional libraries

Certain version of AMS will require different libraries to be installed on the system depending on the MPI library used.

Graphics

In order to run the the graphical user interface (GUI) the computer needs to have an OpenGL-capable graphics subsystem (hardware, drivers and libraries). Besides that, on Linux the following (or equivalent) packages must be installed:

fontconfig
freetype
libdrm
libICE
libSM
libstdc++
libcrypt (part of glibc on many modern systems)
libX11
libXau
libXdmcp
libXext
libXft
libxkbcommon-x11
libXrender
libXScrnSaver (Ubuntu users may need to install libXss1)
libXt
libXxf86vm
mesa-libGL
mesa-libGLU

The GUI will not able to start without shared libraries provided by these packages.

NOTE: If you receive an error about libXss (libXss.so.1: cannot open shared object file: No such file or directory), you need to install libXScrnSaver (redhat/centos: yum install libXScrnSaver) or libxss1 (ubuntu/debian: sudo apt install libxss1).

Compiler

If you have a license for the source code, you can compile the source yourself, with or without your own modifications.

The source consists mainly of Fortran95/2003 code, with some small parts written in C. Some of the Fortran2003 features are also used so a compiler supporting it is required. You must use object-compatible Fortran and C compilers to those we are using on the same platform, since some parts of the code are available only as object modules. For all x86 platforms it is currently Intel OneAPI Fortran 2021.4.0. It is very unlikely that other compilers, or even a different major version of the same compiler, will work with these object modules. We cannot support compilers different from what we are using ourselves.

To check which compiler to use, check the detailed machine information on the Download section of our web site.