Low-power server computing is big right now so big Intel has even begun researching massively parallel low-power servers based on Intel Atom. Yeah baby.
A joint Intel Labs/Carnegie Mellon University project aims to use low-power CPUs to power server computing nodes, transferring certain less CPU-intensive workloads to lower-power processing clusters and freeing up processing power where it is needed.
The project goes by the name of Intel FAWN, where FAWN stands for 'Fast Array of Wimpy Nodes'. No, really; the 'Wimpy Nodes' part being a bunch of Atom-based systems operating in parallel to provide server computing power at a fraction of the power.
Intel researcher Michael Kaminsky, explained that Intel is aiming to reduce power consumption in data-intensive workloads by a factor of three. Speaking to eWeek, Kaminsky underlined that these compute clusters were built with technology you can pick up at any store: a mini-ITX Atom motherboard, memory and an SSD (although we'd like to believe the use of the term SSD includes CF cards).
iFAWN 4 prototype, as witty engineers call it.
Does a massively-parallelised Atom system make for a good server? We have no clue. It depends far too much on the underlying software and how it performs load-balancing. However, it would make for a cheap, power saving set-up.
Intel has a roadmap for massively parallel high-performance computing, one that involves our old friend Larrabee, now reincarnated under the name Knights Corner and Knights Ferry (HPC server and HPC card, respectively). It does seem overkill to use high-power servers such as these for moving chunks of data around, so an Atom-based cluster might be quite useful.
Intel said last week that it won't plug Atom for severs. Mainly, we assume, because an Atom chip is as cheap as, um, chips, while a Xeon costs an ARM and a leg - most of that being margin.