CFD Simulations on a Raspberry Pi  Cluster

James J. Guthrie

17th April, 2015


The release of the second generation Raspberry Pi has made the concept of running a cluster of Raspberry Pi's much more sensible vs. a cluster of first generation Pi's. The University of Southampton built a cluster of 64 Raspberry Pi's. [IridisPi] The Iridis Pi has a peak CPU performance benchmark was around 1 GIGAFLOPS (floating point operations per second)

The cluster presented here is made of 3 second generation Pi's (nodes) and has a peak performance of over 3 GIGAFLOPS.

The first generation Pi has 1 core running at 700 MHz and the 'Model B' used in the Iridis Pi has 512 MB of RAM. The second generation Pi has a quad core CPU running each core at 900 MHz each and 1 GB of RAM. This makes the comparison between Iridis Pi and this cluster: 64 processors vs 12 processors, and 32 GB of RAM vs 3 GB of RAM, meaning the boost in benchmark performance is very exciting.

The Iridis Pi cost around £2,500 excluding networking equipment. This cluster cost £120 excluding networking equipment.

The HPL (High Performance Linpack) benchmark tool is used in this benchmark with a problem size (n) of 17000 resulting in 3.048 GIGAFLOPS. Each Pi is connected to a 100Mb router by ethernet and has an 8GB SD card.

Illustration 1 shows the cluster built in a tower of Mega Blocks. The tower also contains a first generation 'Model B' Pi at the bottom however this is not used in cluster operations. The cluster was tested with all 4 Pi's shown there and the performance was closer to 1 GIGAFLOPS even though it was over 13 processors.

CFD Simulations with OpenFOAM

OpenFOAM 2.3.1 source was downloaded from and was compiled on one of the Pi's then the SD card was duplicated for the other 2 Pi's. OpenFOAM 2.3.1 utilises Open MPI 1.6.5 to communicate across the cluster.

Wmake has to compiled with the -mfloat-abi=hard flag then OpenFOAM can be compiled on the ARM architecture.

The 3 GIGAFLOPS performance means simulations take around the same time to complete as on a single core on a relatively modern workstation.


Future Work

The plans for this cluster are to steadily add more nodes and eventually find the limitations in the current network architecture used. The cluster will be used for the simulations featuring in the author's PhD thesis and the cost will likely be met by the author.



James J. Guthrie, BEng is a PhD researcher at the University of Strathclyde in Glasgow, Scotland. The author's PhD is an investigation into the use of supercritical fluids in internal combustion engines, focusing on supercritical combustion and equation of states.

james dot guthrie at strath dot ac dot uk


IridisPi: Simon J. Cox, James T. Cox, Richard P. Boardman, Steven J. Johnston, Mark Scott, Neil S. O'Brien, Iridis-pi: a low-cost, compact demonstration cluster, 2013.