Software Support Overview

The HKUST HPC clusters provide a comprehensive suite of pre-installed software modules. We utilize Spack for package management and Lmod for environment module management, ensuring a flexible and reproducible research environment.

Our software stack is organized into a three-level hierarchy:

Core (Root): Compilers, Interpreters (Python, R), Utilities, and GPU Drivers.
Compiler Dependent: Libraries and tools built with a specific compiler (e.g., FFTW built with GCC).
MPI Dependent: Parallel applications built with a specific Compiler and MPI combination (e.g., HDF5 built with GCC and OpenMPI).

Note

Access to the latest software requires activating the ‘Edge’ Spack instance. See below for details.

Spack Instances 

We maintain multiple Spack instances to provide different sets of software packages:

Characteristic	Default Instance	Edge Instance
Date of Build	Mid 2024	Weekly updates
Software Ecosystem	Basic	Large, multi-versioned
CUDA Support	✓	✓
Lmod Hierarchy	Flat	Hierarchical (Compiler + MPI)
Spack Version	0.22.0 (May 2024)	0.23.x (Mar 2025) with package backports
Status	Deprecated (Frozen)	Semi-stable
Spack Chaining	✓	⚠ [1]
Compatibility	Standalone	Forward-compatible
$SPACK_ROOT	HPC4: `/opt/shared/spack` Superpod: ✗	HPC4: `/opt/shared/.spack-edge` Superpod: `/scratch/spack/2025`

Tip

We recommend using the edge instance for all new work as it provides the latest software packages and is forward-compatible with future updates.

The default instance is deprecated and maintained only for backward compatibility.

Activating a Spack Instance 

The default instance, when available, is automatically activated upon login on HPC4.

To activate the alternative instance, source its setup script:

# Acknowledge the Terms of Service and activate an alternative spack instance
source "${SPACK_ROOT}/dist/bin/setup-env.sh"

Note

The first-time activation includes a bootstrapping process that may take a few minutes. Subsequent activations will be much faster.

When activating an alternative Spack instance for the first time, you’ll see similar output to below bootstrapping the Spack instance.

Note

It is normal to have some warnings during bootstrap shown below, please ignore these warnings as they do not affect the bootstrap process.

Warning: error detecting "msvc" ...
==> Warning: The default behavior of tarfile extraction ...
==> Warning: lib/libgfortran.so.5
        libz.so.1 => not found

Example Activation Output

[user@login2 ~]$ source /opt/shared/.spack-edge/dist/bin/setup-env.sh

You are using non-default spack instances.
Please do not mix packages installed from different spack instances.

This script will unload all other spack instances and modules automatically.

==> Checking spack config and cache paths


    Spack instance root: /opt/shared/.spack-edge
    Shared apps and modules: /opt/shared/.spack-edge/dist
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    Your spack config: /home/user/.spack-edge
    Your spack apps: /home/user/.spack-edge
    Tmpdir (TMPDIR and TMP): /tmp/user-XXXX

==> Activate spack instance [edge]? [y/N] y

i=> You can use '-y' option to skip this confirmation next time.

==> First launch: bootstrapping spack [edge]
    This may take a few minutes, please wait...
==> First launch: bootstrapping spack [dev]
    This may take a few minutes, please wait...
==> Warning: error detecting "msvc" from prefix /opt/shared/.spack-dev/opt/spack/linux-rocky9-x86_64_v4/gcc-11.5.0.spack/intel-oneapi-compilers-2025.0.4-sn26au2eyxigpsati3gb5oxmtku6s5uo/compiler/2025.0/bin: [KeyError: 'cxx']
==> Fetching https://ghcr.io/v2/spack/bootstrap-buildcache-v1/blobs/sha256:16f27dafb233a9aff14adff75112d0a8b2e03492f796f12101785c02950aa7f6
==> Fetching https://ghcr.io/v2/spack/bootstrap-buildcache-v1/blobs/sha256:b76a4eaef54b24d0ea9b8dfa9392f7ab519f918ae7e1a8bb919539d9adeddbcb
==> Installing "gcc-runtime@=10.2.1 build_system=generic arch=linux-centos7-x86_64 %gcc@=10.2.1" from a buildcache
==> Warning: The default behavior of tarfile extraction has been changed to disallow common exploits (including CVE-2007-4559). By default, absolute/parent paths are disallowed and some mode bits are cleared. See https://access.redhat.com/articles/7004769 for more details.
==> Warning: lib/libgfortran.so.5
        libz.so.1 => not found
dependencies:
==> Fetching https://ghcr.io/v2/spack/bootstrap-buildcache-v1/blobs/sha256:5367a0d038a87532fbbe373da31502bd32e399cc113343f403ebbc9d8ca7d552
==> Fetching https://ghcr.io/v2/spack/bootstrap-buildcache-v1/blobs/sha256:79dfb7064e7993a97474c5f6b7560254fe19465a6c4cfc44569852e5a6ab542b
==> Installing "patchelf@=0.17.2 build_system=autotools arch=linux-centos7-x86_64 %gcc@=10.2.1" from a buildcache
==> Fetching https://ghcr.io/v2/spack/bootstrap-buildcache-v1/blobs/sha256:82ec278bef26c42303a2c2c888612c0d37babef615bc9a0003530e0b8b4d3d2c
==> Fetching https://ghcr.io/v2/spack/bootstrap-buildcache-v1/blobs/sha256:0c5831932608e7b4084fc6ce60e2b67b77dab76e5515303a049d4d30cd772321
==> Installing "clingo-bootstrap@=spack~docs+ipo+optimized+python+static_libstdcpp build_system=cmake build_type=Release generator=make patches=bebb819,ec99431 arch=linux-centos7-x86_64 %gcc@=10.2.1" from a buildcache


==> Spack [edge] environment is ready

For non-interactive shells (e.g., in batch scripts), use the -y flag to avoid prompts

# Non-interactive activation, do not ask for input
source /opt/shared/.spack-edge/dist/bin/setup-env.sh -y

After sourcing, you can verify the active instance:

# Check which Spack instance is active
echo $SPACK_VARIANT

Caution

Do not add Spack activation to your ~/.bashrc or ~/.bash_profile. Activating Spack automatically may cause slow login and compatibility issues with SLURM batch jobs.

Best practice: Activate Spack on-demand when you need to use software modules, or at the beginning of SLURM job scripts. For batch scripts, always use the -y flag to skip confirmation prompts:

# In SLURM job scripts
source /opt/shared/.spack-edge/dist/bin/setup-env.sh -y
module load python/3.12.9
# ... rest of your job script

Spack Environment Variables 

When a Spack instance is activated, we configure Spack to use certain paths and set a number of environment variables.

Variable	Description
`$SPACK_ROOT`	Root directory of the active Spack instance
`$SPACK_VARIANT`	Name of the active instance
`$SPACK_PYTHON`	Python interpreter used by Spack
`$SPACK_SYSTEM_CONFIG_PATH`	System-level Spack configuration directory
`$SPACK_USER_CONFIG_PATH`	User-specific Spack configuration directory
`$SPACK_USER_CACHE_PATH`	User-specific Spack cache directory

Lmod Modules 

Lmod modules are provided by the Spack instance, so the available modules depend on which instance is active.

Tip

Check that your desired Spack instance is active before using module commands.

Loading and managing modules 

To load a specific software module, use the module command:

# List available modules
module avail

# Search for a specific module across all levels
module spider netcdf-c

# Get detailed information about a specific module
#module spider python/<version>-<hash>

# Load a specific module (e.g., Python 3.12)
module load python/3.12

# Unload a specific module
module unload python

# Unload all modules
module purge

Important

Do not include the 7-digit hash suffix when loading modules (e.g., use python/3.12.9 instead of python/3.12.9-abc1234). The hash may change when packages are rebuilt, which would break your scripts.

Understanding the Lmod Hierarchy 

We adopt a hierarchical module system where loading a compiler or MPI library reveals additional modules built with that toolchain.

Example workflow:

Initially, you see only core modules:

$ module avail
---- Core Modules ----
gcc/14.2.0
intel-oneapi-compilers/2025.0.4
python/3.12.9
...

After loading a compiler, compiler-dependent modules become available:

$ module load intel-oneapi-compilers/2025.0.4
$ module avail

---- Core Modules ----
...

---- /opt/shared/.spack-edge/share/spack/lmod/linux-rocky9-x86_64/oneapi/2025.0.4 ----
fftw/3.3.10                   openmpi/4.1.8
intel-oneapi-mpi/2021.14.2    openmpi/5.0.6              (D)
...

After loading an MPI library, MPI-dependent modules appear:

$ module load openmpi/5.0.6
$ module avail

---- Core Modules ----
...

---- /opt/shared/.spack-edge/.../oneapi/2025.0.4 ----
...

---- /opt/shared/.spack-edge/.../oneapi/2025.0.4/openmpi/5.0.6 ----
hdf5/1.14.5            netcdf-c/4.9.2
parallel-netcdf/1.12.3 scalapack/2.2.0
...

This hierarchy ensures you’re using compatible versions of libraries built with the same compiler and MPI combination.

SLURM Integration 

When submitting batch jobs to SLURM, a typical workflow is to

activate the Spack instance non-interactively
load the required modules
continue your program settings and run your program

This example demonstrates the hierarchical module system: first load the compiler, then the MPI implementation, and finally any MPI-dependent libraries like NetCDF.

#!/bin/bash

##SBATCH ...

# Activate Spack instance (use -y flag to skip prompts)
export SPACK_ROOT="/path/to/your/selected/spack/instance"
source "${SPACK_ROOT}/dist/bin/setup-env.sh" -y

# Load compiler and MPI
module load intel-oneapi-compilers/2025.0.4
module load intel-oneapi-mpi/2021.14.2

# Load application libraries
module load netcdf-c/4.9.2

# Set up your application related settings
#...
# Run your MPI program
#srun my_mpi_program

Compiler Support 

Here is a list of maintained C/C++/Fortran compilers for building high-performance applications.

Danger

Architecture Compatibility Warning

Libraries built with the AMD AOCC compiler are strictly optimized for the AMD Zen4 architecture.

Do not load AOCC modules on Intel nodes.
Do not submit jobs using AOCC-compiled code to Intel partitions.

Doing so will result in Illegal Instruction errors and immediate job failure.

Note

Bold versions are the default versions loaded when using module load without specifying a version number.

Deprecated packages are marked as versions @edge in the edge Spack variant, deprecated packages will be removed in the next production variant.

Compiler	Versions	Library Support	MPI Support	Compiler Command
AMD AOCC	4.2.0 @edge 5.0.0	✓	✓	CC=clang CXX=clang++ FC=flang
GNU gcc	12.4.0 13.3.0 14.2.0			CC=gcc CXX=g++ FC=gfortran
Intel OneAPI	2021.4.0 @edge 2022.2.1 @edge 2023.2.4 2024.2.1 2025.0.4	✓	✓	CC=icx CXX=icpx FC=ifx
Intel Classic	2021.10.0	✓	✓	CC=icc CXX=icpc FC=ifort
Nvidia nvhpc	23.11 24.11 25.1			CC=nvc CXX=nvc++ FC=nvfortran

MPI Support 

The following table shows tested compiler and MPI combinations available in the edge instance. A checkmark (✓) indicates the combination has been tested and is available.

Compiler / MPI	Intel MPI 2021.14	OpenMPI 4.1.x	OpenMPI 5.x
AOCC 5.0		✓	✓
AOCC 4.2
Intel OneAPI 2025.x	✓	✓	✓
Intel OneAPI 2024.x	✓	✓	✓
Intel OneAPI 2023.x	✓	✓	✓
Intel OneAPI 2022.x
Intel OneAPI 2021.x
Intel Classic 2021.10 (OneAPI 2023)	✓
NVHPC 25.1		✓	✓
NVHPC 24.11		✓	✓
NVHPC 23.11		✓	✓
GCC 14.x
GCC 13.x
GCC 12.x

CUDA Support 

The following table shows which compilers have CUDA support enabled for GPU programming and their MPI compatibility when building packages via Spack.

Warning

[NOT tested] For compilers without Spack’s native CUDA support (GCC, AOCC), directly loading CUDA Toolkit separately with module load cuda might work.

Compiler / MPI	CUDA Support	Intel MPI 2021.14	OpenMPI 4.1.x	OpenMPI 5.x
NVHPC 25.1	✓		✓	✓
NVHPC 24.11	✓		✓	✓
NVHPC 23.11	✓		✓	✓
Intel OneAPI 2025	✓	✓	✓	✓
Intel OneAPI 2024	✓	✓	✓	✓
Intel OneAPI 2023	✓	✓	✓	✓
Intel OneAPI 2022	✗
Intel OneAPI 2021	✗
Intel Classic 2021.10	✗
GCC (all versions)	✗
AOCC (all versions)	✗

List of pre-built software and libraries 

Libraries compiled by Intel compilers are optimized for latest Intel and AMD CPUs, while those compiled by AOCC are optimized for AMD CPUs only.

Danger

Architecture Compatibility Warning

Libraries built with the AMD AOCC compiler are strictly optimized for the AMD Zen4 architecture.

Do not load AOCC modules on Intel nodes.
Do not submit jobs using AOCC-compiled code to Intel partitions.

Doing so will result in Illegal Instruction errors and immediate job failure.

Note

Bold versions are the default versions loaded when using module load without specifying a version number.

Deprecated packages are marked as versions @edge in the edge Spack variant, deprecated packages will be removed in the next production variant.

HPC Applications 

These applications are scientific software that is built for general use. They may have both serial and parallel versions depending on their nature.

Software	Description	Version	Compiler / MPI	GPU
lammps	Large-scale Atomic/Molecular Massively Parallel Simulator	20250204	- Intel OneAPI 2025 - Intel MPI 2021.14	✓
mpas-model	Model for Prediction Across Scales (Atmosphere/Ocean)	7.3 8.0.2 8.1.0	- Intel OneAPI 2025 - Intel MPI 2021.14
openfoam-org	OpenFOAM CFD Toolbox (org version)	11 12	- Intel OneAPI 2025 - Intel MPI 2021.14

MPI-enabled Libraries 

These libraries become available after loading both a compiler and an MPI implementation. They are built specifically for parallel computing applications.

Software	Description	Version	Docs
fftw	Fastest Fourier Transform in the West - Single & Double precision	3.3.10
hdf5	Hierarchical Data Format 5 - Subfile, szip enabled	1.14.5
netcdf-c	Network Common Data Form (C library) - Pnetcdf, szip enabled	4.9.2	Details
netcdf-fortran	Network Common Data Form (Fortran library)	4.6.1	Details
parallel-netcdf	Parallel I/O library for NetCDF file access	1.14.0	Details
parallelio	High-level parallel I/O library - Pnetcdf enabled	2.6.3

Intel Compiler Specific Libraries 

Libraries exclusively available when using Intel OneAPI or Intel Classic compilers with MPI.

Software	Description	Version	Docs
intel-oneapi-mkl	Intel Math Kernel Library - ScaLAPACK, Distributed FFT etc. included	2023.2.0 2024.2.2 2025.0.1

Compiler-optimized Libraries 

These libraries are optimized for specific compiler toolchains and become available after loading the corresponding compiler module.

Software	Description	Version	Docs
boost	Boost C++ Libraries - all subsets included	1.87.0
eigen	C++ Template Library for Linear Algebra	3.4.0
fftw	Fastest Fourier Transform in the West - Single & Double precision	3.3.10
glib	GNOME Core Library	2.72.4
gsl	GNU Scientific Library	2.8
libxc	Library of Exchange and Correlation Functionals	7.0.0	Details

AOCC Specific Libraries 

AMD-optimized libraries available exclusively with AMD Optimizing C/C++ Compiler (AOCC).

Software	Description	Version
AMD AOCL	AMD AOCL Libraries - all features enabled	5.0
amdblis	AMD BLAS-Like Instant Software (Optimized BLAS)	5.0
amdfftw	AMD Optimized FFTW (Fast Fourier Transform)	5.0
amdlibflame	AMD FLAME Library (Linear algebra)	5.0
amdlibm	AMD Math Library (Optimized libm)	5.0

Generic Software 

The following sections list categories of software packages available at the Root level. Packages are compiled with a freshly built GCC 11.5.0 compiler independent of OS.

GPU & AI Infrastructure 

Core libraries and drivers for GPU computing.

Software	Description	Version	Docs
CUDA	NVIDIA Compute Unified Device Architecture	11.8.0 12.2.2 12.3.2 12.4.1 12.5.1 12.6.3 12.8.0
cuDNN	NVIDIA CUDA Deep Neural Network library for CUDA 11/12	8.9.7 9.2.0

Math & Data Science 

Platforms for mathematical computing, statistical analysis, and data science.

Software	Description	Version	Docs
MATLAB	Numerical computing platform	R2019b R2022b R2023b	Details
Octave	Scientific Programming Language (MATLAB-compatible)	9.4.0
Python	Python Programming Language	3.9.21 3.10.16 3.11.11 3.12.9 3.13.2	Details
Anaconda3	Python Data Science Platform	2024.10	Details
R	R Language for Statistical Computing	4.4.2	Details
RStudio	Integrated Development Environment for R	2024.12.1	Details

Language Runtimes 

Runtime environments for various programming languages.

Software	Description	Version	Docs
OpenJDK	Open Source Java Development Kit	1.8.0_265 11.0.23 17.0.11	Details
Rust	Rust Systems Programming Language	1.85.0
Go	The Go Programming Language	1.24.1

Visualization & Media 

Tools for graphics, visualization, and media processing.

Software	Description	Version
Gnuplot	Portable command-line driven graphing utility	6.0.0
VMD	Visual Molecular Dynamics	1.9.3
ImageMagick	Image creation and modification tools	7.1.1-39
FFmpeg	Video and Audio processing suite	4.4.4 5.1.4 6.1.1 7.1

Libraries 

Core system libraries and development tools.

Software	Description	Version
Jasper	JPEG-2000 codec library	1.900.31 4.2.8
libpng	PNG reference library	1.6.39
libtirpc	Transport Independent RPC library	1.3.3

Build Tools & Development 

Tools for building, configuring, and managing software projects.

Software	Description	Version
Binutils	GNU Binary Utilities	2.43.1
CMake	Cross-platform build system generator	3.31.6
Meson	Modern build system	1.7.0
Ninja	Small build system with a focus on speed	1.12.1
Gmake	GNU Make	4.4.1

Git Version Control 

Version control tools for managing code repositories and large files.

Software	Description	Version	Docs
Git	Distributed version control system	2.48.1
Git-LFS	Git Large File Storage	3.5.1

Other Packages 

Scripting Languages: High-level scripting and interpreted languages.
Other Version Control Systems: Alternative version control systems.
Debugging & Profiling: Tools for debugging, testing, and code analysis.
Utilities & Tools: General system utilities, editors, and productivity tools.
Cloud CLI: Command-line tools for cloud services.
Package Managers: Package management tools.
Additional Build Tools: Less commonly used build tools and Java toolchains.
Additional Language Runtimes: Additional programming language runtimes.

Scripting Languages

Software	Description	Version	Docs
Perl	Practical Extraction and Report Language	5.40.0	Details
Ruby	Dynamic, open source programming language	3.3.5
Lua	Lightweight, multi-paradigm scripting language	5.4.6
Tcl	Tool Command Language	8.6.12
Bash	GNU Bourne Again Shell	5.2
Tcsh	C Shell with file name completion	6.24.14

Other Version Control Systems

Software	Description	Version
Subversion	Centralized version control system	1.14.2
Mercurial	Distributed version control system	6.7.3
CVS	Concurrent Versions System	1.12.13

Utilities & Tools

Software	Description	Version
Emacs	Extensible text editor	30.1
Neovim	Hyperextensible Vim-based text editor	0.11.5
Zsh	Z Shell	5.9
Screen	Terminal multiplexer	4.9.1
Parallel	GNU Parallel (Shell tool for executing jobs in parallel)	20240822
dos2unix	Text file format converter	7.4.4
EasyBuild	Software build and installation framework	4.7.0

Additional Build Tools

Software	Description	Version
Autotools	GNU Build System (Autoconf, Automake, Libtool)	master
SCons	Software construction tool	4.7.0
Bazel	Build and test tool (Google)	7.0.2
Maven	Build automation tool for Java	3.9.8
Gradle	Build automation tool for multi-language	8.10.2
Ant	Java build tool	1.10.14

Additional Language Runtimes

Software	Description	Version
Dotnet Core SDK	.NET Core Software Development Kit	8.0.4
Node.js	JavaScript runtime built on Chrome’s V8	22.14.0
OCaml	Functional programming language	5.2.1

Software	Description	Version	Docs
GDB	GNU Debugger	15.2
GoogleTest	Google C++ Testing Framework	1.15.2

Software	Description	Version	Docs
AWS CLI v2	Amazon Web Services Command Line Interface	2.24.24
Google Cloud CLI	Google Cloud SDK	504.0.1

Software	Description	Version
NPM	Node Package Manager	11.2.0
Yarn	JavaScript package manager	1.22.22
FPM	Fortran Package Management	0.10.0