Intel Toolchains

A toolchain consists of one or more compilers, usually put together with some libraries for specific functionality, e.g., for using an MPI stack for distributed computing, or to provide optimized routines for commonly used maths operations, e.g., the well-known BLAS/LAPACK APIs for linear algebra routines.

Hence loading a toolchain modulefile will load other modulefiles (and make its own settings) to give you a compiler, and a set of libraries compiled with that compiler, that you can then use to build your own applications – for example compiling your own source code. Many applications require the various packages provided by the toolchain, so using a toolchain gives a consistent method of building applications.

icc/ifort/icpc tools

The following Intel-compiler toolchains are available. The version number (e.g., 2020.02) refers to the version of the Intel Compiler. You should load one of the following modulefiles:

# Intel C, C++ and Fortran Compilers + Intel Math Kernel Libraries (MKL) + OpenMPI (parallel library) 
module load iomkl/2020.02           # Compilers+MKL at 2020.2.254, openmpi 4.0.4

# Intel C, C++ and Fortran Compilers + OpenMPI (parallel library)
module load iompi/2020.02           # Compilers at 2020.2.254, openmpi 4.0.4

# Intel C, C++ and Fortran Compilers + Intel Math Kernel Libraries (MKL)
module load iimkl/2020.02           # Compilers+MKL at 2020.2.254 + access to some gcc 9.3.0 core libs

# Intel C, C++, Fortran Compilers (not a "toolchain", just the compilers)
module load iccifort/2020.2.254     # Compilers at 2020.2.254 + access to some gcc 9.3.0 core libs

You may wish to use the iompi (no MKL) from the list above if using some other library for BLAS or FFTW routines.

OneAPI tools

Alternatively, the Intel OneAPI suite is available, which provides newer versions of the Intel compilers, libraries and tools:

• OneAPI 2025.3.1 helper modulefile:

  # Intel OneAPI 2025.3.1 helper - load this, then follow the instructions:
  module load compilers/oneapi/2025.3.1
    #
    # You can now load the modulefiles for the OneAPI components you require:
    # If you're on the login node, it will tell you to then load further modulefiles:
    module load umf compiler-rt tbb compiler
    #
    # Or, if you want to use VTune:
    module load vtune
  
    # Or alternatively, you can now source the traditional Intel setup scripts:
    source $ONEAPIDIR/setvars.sh
  

• OneAPI 2025.0.1 helper modulefile:

  # Intel OneAPI 2025.0.1 helper - load this, then follow the instructions:
  module load compilers/oneapi/2025.0.1
    #
    # You can now load the modulefiles for the OneAPI components you require:
    # If you're on the login node, it will tell you to then load further modulefiles:
    module load umf compiler-rt tbb compiler
    #
    # Or, if you want to use VTune:
    module load vtune
  
    # Or alternatively, you can now source the traditional Intel setup scripts:
    source $ONEAPIDIR/setvars.sh
  

• OneAPI 2024.2.0 helper modulefile:

  # Intel OneAPI 2024.2 helper - load this, then follow the instructions:
  module load compilers/oneapi/2024.2.0
    #
    # You can now load the modulefiles for the OneAPI components you require:
    # If you're on the login node, it will tell you to then load further modulefiles:
    module load compiler-rt tbb compiler
    #
    # Or, if you want to use VTune:
    module load vtune
  
    # Or alternatively, you can now source the traditional Intel setup scripts:
    source $ONEAPIDIR/setvars.sh
  

If you are submitting jobscripts, you should load the modulefiles that you would have loading on the login node:

#!/bin/bash --login
#SBATCH -p multicore
#SBATCH -n 8
module load compilers/oneapi/2025.0.1
module load umf compiler-rt tbb compiler/2025.0.1
./my_oneapi_app.exe 

Compiler Flags

See elsewhere in the documentation for details of the Intel Compiler, the MKL and OpenMPI. For now, when compiling code with the Intel compiler, we recommend the following flags to ensure your app runs on the CSF4 Cascade Lake CPUs and also any nodes from the CSF3 that we may move in to CSF4 (there are no plans to do this at the moment.)

# Flags to add to icc / icpc / ifort (e.g., vi CFLAGS etc in your Makefile)
-O3 -xAVX -axCORE-AVX512,CORE-AVX2,AVX

OpenMPI

OpenMPI compiled with the Intel compiler (various versions) is available via OpenMPI modulefiles. In some cases, these will load the necessary compiler modulefiles for you:

The choice of modulefile here depends on which FORTRAN compiler you wish to compile your MPI code with.

• If you want to compile FORTRAN code with the Intel ifx compiler:

  # Note: mpif90, mpif77 and mpifort all use the 'ifx' Intel fortran compiler, not 'ifort'
  module load mpi/oneapi-2024.2/openmpi/4.1.8
  
  # If you require the MKL, you must now load:
  module load mkl/2024.2
  

• If you want to compiler FORTRAN code with the traditional Intel ifort compiler:

  # Note: mpif90, mpif77 and mpifort all use the 'ifort' Intel fortran compiler, not 'ifx'
  # You will also need to load extra modulefiles afterwards as shown:
  module load mpi/intel-2024.2/openmpi/4.1.8
  module load compiler-rt tbb compiler
  
  # If you require the MKL, you must now load:
  module load mkl/2024.2
  

In both cases above, the OneAPI C and C++ compilers will be used: the C compiler will be icx and the C++ compiler will be icpx.