Cheap Supercomputers: LANL has 750-node Raspberry Pi Development Clusters
One of the more esoteric announcements to come out of SuperComputing 17, an annual conference on high-performance computing, is that one of the largest US scientific institutions is investing in Raspberry Pi-based clusters to aid in development work. The Los Alamos National Laboratory's [LANL] High Performance Computing Division now has access to 750-node Raspberry Pi clusters as part of the first step towards a development program to assist in programming much larger machines.
The platform at LANL leverages a modular cluster design from BitScope Designs, with five rack-mount Bitscope Cluster Modules, each with 150 Raspberry Pi boards with integrated network switches. With each of the 750 chips packing four cores, it offers a 3000-core highly parallelizable platform that emulates an ARM-based supercomputer, allowing researchers to test development code without requiring a power-hungry machine at significant cost to the taxpayer. The full 750-node cluster, running 2-3 W per processor, runs at 1000W idle, 3000W at typical and 4000W at peak (with the switches) and is substantially cheaper, if also computationally a lot slower. After development using the Pi clusters, frameworks can then be ported to the larger scale supercomputers available at LANL, such as Trinity and Crossroads.
(Score: 2) by stormwyrm on Thursday November 16 2017, @01:26AM (1 child)
You'd need some really fast interconnects for most serious supercomputing applications. The Raspberry Pi 3 has a built-in 10/100 Mbps Ethernet, which is pretty damn slow for as far as that goes. There's 802.11n wireless on the Pi 3 too, but I seriously doubt that you can really get the nominal maximum of 150 Mbps reliably. Unless they have a custom USB 2.0 protocol that could do 480 Mbps, but it doesn't look like it. In contrast, Infiniband, widely used for supercomputing interconnections, does at least 2 Gbps and can go into the 100+ Gbps range. I'd have tried using the ODROID C2 instead, which at least has an honest to goodness gigabit Ethernet and isn't that much more expensive.
Numquam ponenda est pluralitas sine necessitate.
(Score: 1) by kai_h on Thursday November 16 2017, @08:50AM
This cluster is not designed to give you the performance of a supercomputer, but rather as a test bed to run code on to see how well it parallelises on a cluster. It's like a cluster with it's L Plates on.
Processing speed & interconnect speed are secondary considerations over allowing researchers to see how their code handles a large number of nodes.