In late 2010 ago I noted that startup SeaMicro had introduced an ultra-dense server using Intel Atom chips in an innovative fabric-based architecture that allowed them to factor out much of the power overhead from a large multi-CPU server..
Along with many observers, I noted that the original SeaMicro server was well-suited to many light-weight edge processing tasks, but that the system would not support more traditional compute-intensive tasks due to the performance of the Atom core. I was, however, quite taken with the basic architecture, which uses a proprietary high-speed (1.28 Tb/s) 3D mesh interconnect to allow the CPU cores to share network, BIOS and disk resources that are normally replicated on eper server in conventional designs, with commensurate reductions in power and an increase in density.
18 months later SeaMicro, again with support from Intel, has extended this architecture to the Xeon CPU, and has introduced the SM1000-EX, based on the Intel Xeon E3-1260L, a 45W quad-core design targeted at low power single socket servers, currently widely deployed in HPC environments. The resulting system can be configured with up to 64 CPUs, for a total of 256 cores in 10U.
While impressive, this is not actually leading-edge density - there are other dense-pack servers that can deliver equal or higher core density with Xeon CPUs. What is significant is the power consumption - SeaMicro manages to do this on an aggregate power budget that comes out to a claimed 55W per core, only 10W over the power required for the core itself. In contrast, a conventional server design today would require at least 2 - 3X the power of the core itself, giving the New SeaMicro design a substantial edge in energy efficiency.
For Intel, SeaMicro represents a microscopic proportion of their business, but is a highly visible ally in the quest to prove that x86 servers can compete in an increasingly energy-conscious world as alternative architectures such as ARM raise the spectre of another competing architecture in the datacentre.
For the industry as a whole, I think the implications of some of the recent server developments from Intel/SeaMicro and Calxeda/HP are even more significant - if we connect the dots, we can see the developing shape of what may be the most significant evolution in fundamental server architecture in about 25 years - the transition to scalable fabric-based as opposed to bus-based architectures.
But in the best tradition of Charles Dickens, that’s a story for the next edition :)
Posted by Richard Fichera