More blogs in Arm Community blogs
Tags
Actions
Related blog posts
Related forum threads

Transmogrify the Personal Computer using efficient low power SoC solutions

Andrew N. Sloss
3 minute read time.
Wouldn't it be great if I could transmogrify the PC to become an intelligent network of specialized components? The goal would be to create a much more efficient computation system marrying critical software with specialized System-on-a-Chip (SoC) based hardware. 

Background
About 8 years ago there was a craze to build Beowulf computer class clusters  (concept of building High Performance Computers using standard commodity PCs), so I went about creating a small cluster of PCs and connecting them via Ethernet at home.  Finally after completing the project I switched the system on, the lights flickered due to the huge power surge and it became instantly obvious to me that a new approach was required. Not just for High Performance Computing but for the fundamental design of the PC. The PC has too many standard parts which were cheap to manufacture in the short-term, but are wrong for long-term cost and eco friendliness.

A new way forward
Today we buy a PC from a store and it is a general purpose device. Every 3 or 4 years we buy another PC, even though a lot of the standard components are the same between machines. We have been trained to continue this cycle. Recently, there have been signs that the increases in Instruction Level Parallelism (ILP) is starting to slow down. The technology direction is moving towards energy efficient many-core configurations.  If we go down this route then we can redesign the base architecture of the PC and try to do something new. I want a computer which is composed of specialized computational elements connected via an extremely fast bus network e.g. over fiber optics. This would remove the requirement for complete replacement every few years and allow someone to add components for specialized activities.

For instance, if I use Microsoft Office as my main editing tool I would purchase a Microsoft Office hardware component which is based on a computation engine (i.e. specialized optimized SoC) carefully designed to execute the Office functions efficiently rather than relying on the traditional general purpose CPU designs. The database and filing system reside somewhere else on the system (i.e. locally or in the cloud). This would allow processing to be achieved in a much more distributed and efficient manner. Computation components not in use can be power down. For example when I'm playing a Game then the Office hardware maybe powered-down but my specialized gaming physics engine is most likely powered-up. Even the phone can be part of the system by providing voice and some computational characteristics.

Today we use the same mallet for every problem even when the mallet isn't the right size or weight. Is it time to optimize the hardware architecture and move away from the traditional PC design created in the 1980's? Somehow a method of piping all the various computations into one or more graphical windows would be required to emulate the standard desktops found today. I suspect this would require some clever GPU design for multi computational rendering. I'll leave this to GPU folks.

Eco factor
All of this would be a good challenge for the computing industry which has relied on this old concept for far too long. The interesting challenge of this approach is cost. Does the fast fiber optics and serialization of the physical communication parts make this concept cost prohibitive? I personally would like to see more interesting and alternative computer system architectures where I can build unique ultra low power + high efficient SoC based computer systems at home using standard interconnects. These systems would avoid the repeated cost of buying the same standard components with every new computer generation. It would allow me to have a componentized computer.

Do you think we are ready?
Anonymous
Parents
  • Awesome! I hadn't considered the quantum registers! For me it is more about the distributed computing model approach rather than the general PC computing model but you're right the transport layer would have to be complex, not sure it would require quantum registers but a more practical fiber-optic interconnect.
Comment
  • Awesome! I hadn't considered the quantum registers! For me it is more about the distributed computing model approach rather than the general PC computing model but you're right the transport layer would have to be complex, not sure it would require quantum registers but a more practical fiber-optic interconnect.
Children
No Data
Internet of Things (IoT) blog