Where Servers Consolidate, Clients Diversify

By Tsahi Levent-Levi

Categories: Technology

November 12th, 2009

In case you haven’t noticed the latest trend, our clients/terminals/endpoints are becoming diversified in almost every aspect, while our servers are being consolidated.

I’ve written here more than once about the way that chips and platforms for consumer electronic devices and video conferencing terminals are going towards multiple cores, but where each core has different capabilities: you have a host processor, and a bunch of accelerators for video processing and graphics, encryption, baseband functionality for cellular and wireless, etc.

We’re moving towards a world where a computing device has so much capabilities packed into it, that specialization is important. Just look at the iPhone 3GS or the latest specs of the Motorola Droid.

Looking at the other end, in the server market, there is a move towards packing as many cores/machines/processors as possible into a limited real-estate. Up until now, my understanding was that you simply take as many x86 processors as possible, place them on them motherboards, pack them up and ship them in a container to some data center.

This was until I bumped into hoopoe, a company offering cloud computing over GPUs instead of CPUs. While you might say it provides a different service than that of usual cloud computing, it still goes to the same thinking of packing as many “same-processing-power” as possible into as little space as possible.

To me it all means two simple things:

• Cloud computing, which is definitely growing, is going to take form of a large farm of “flat” processing devices. You’ll be able to go choose a cloud service that fits your needs, not only from an SLA point of view but also from the type of processing you are looking for – CPUs, GPUs, whatever. There will be a farm for every different kind of a job.
• End-user devices will diversify, providing a lot more than a single core, where some cores might be radically different than the rest of the cores, making sure specific tasks get the special attention they are looking for. It’s the end of Moore’s Law as we knew it, as we will move towards designing our hardware to meet the needs of our application.