SPM/Faster SPM

SPM Optimisation

Adjusting SPM settings

A defaults variable called maxmem indicates how much memory can be used at the same time during GLM estimation. If your computer has a large amount of RAM, you can increase that memory setting in spm_defaults.m:

defaults.stats.maxmem   = 2^30;

• 2^32 = 4GB
• 2^31 = 2GB
• 2^30 = 1GB
• 2^29 = 512MB

In SPM12, there is another defaults variable called resmem governing whether temporary files during GLM estimation are stored on disk (false) or kept in memory (true). If you have enough available RAM, not writing the files to disk will speed the estimation.

defaults.stats.resmem   = true;

Compiling the MEX files

The compiled MEX files provided with SPM are built in such a way to be compatible with most platforms and MATLAB versions but you might benefit from compiling them for your exact platform/MATLAB version - some C compilers might also produce better optimised binaries (such as Intel Compilers). See the installation pages for more details on how to recompile SPM MEX files.

MATLAB Optimisation

General statements

Maximizing MATLAB Performance.

Note that it is not recommended to disable the JAVA Virtual Machine when launching MATLAB (matlab -nojvm). If you don't want to use the MATLAB desktop, you can preferably launch MATLAB with:

matlab -nodesktop

When using Matlab in 'nodesktop' mode, initialise SPM in the following manner to prevent graphics windows from opening:

spm('defaults', 'fmri')
spm_jobman('initcfg')
spm_get_defaults('cmdline',true)

(Substituting 'fmri' for 'pet' or 'eeg' as appropriate.)

Multithreading

Recent MATLABs support implicit multiprocessing allowing to run multiple threads on a single machine without any change to the MATLAB code itself: this requires a multiple CPU (multiprocessor or multicore) system. The gain in compute time with SPM is not dramatic though.

See:

• https://www.mathworks.com/help/matlab/ref/maxnumcompthreads.html

If you run many MATLAB sessions in parallel to manually distribute your SPM processings, it is recommended to set the number of computational threads to one.

Updating BLAS/LAPACK

Install the latest Basic Linear Algebra Subroutines (BLAS)/Linear Algebra PACKage (LAPACK) for your system, as MATLAB doesn't ship the latest version in its releases.

• BLAS
• LAPACK

The main choices are between:

• ATLAS - Automatically Tuned Linear Algebra Software (open source)
• MKL - Intel Math Kernel Library
• ACML - AMD Core Math Library
• GotoBLAS - GotoBLAS (Texas Advanced Computing Center)

For more information on how to upgrade your MATLAB libraries with ATLAS/MKL, see:

• http://software.intel.com/en-us/articles/using-intel-mkl-with-matlab/
• http://software.intel.com/en-us/articles/intel-math-kernel-library-intel-mkl-for-windows-using-intel-mkl-in-matlab-executable-mex-files/
• http://imaging.mrc-cbu.cam.ac.uk/imaging/SpmWithPentium4

Parallel Computing

SPM toolboxes

See pSPM for SPM2.

General tools

There is currently work in progress to provide an official parallel/distributed version of SPM.

• MATLAB Parallel Survey
• Parallel MATLAB: Doing it right
• Parallel MATLAB: The next generation

See also the Sun Grid Engine Project (SGE):

• Sun Grid Engine
• qsubfunc

Using the Graphics Processing Unit (GPU)

It is possible for MATLAB to take advantage of the GPU (the processor of the graphic card, by opposition to the CPU), to perform some operations, resulting in significant speed improvements. A number of toolboxes are available:

• Parallel Computing Toolbox GPU Computing with MATLAB
• MATLAB GPU Computing
• MATLAB Acceleration with CUDA-enabled GPUs
• NVIDIA GPU and MATLAB
• NVIDIA NGC
• CUDA SPM GPU-Accelerated SPM image registration
• Deep Learning Frameworks

See also:

• https://www.frontiersin.org/articles/10.3389/fninf.2014.00024/full
• https://www.sciencedirect.com/science/article/pii/S0169260711001957
• https://www.sciencedirect.com/science/article/pii/S1361841513000820
• https://link.springer.com/article/10.3758%2Fs13415-013-0165-7
• https://doi.org/10.1109/MSP.2009.935387