[ 074 ]
FEATURERISE OF THE ACCELERATORSEVER THOUGHT ABOUT the range of
computing devices in your home? It could be
any of desktop PCs, laptops, games consoles,
smartphones, tablets, media players, set-top
boxes, smart TVs and the list goes on. What
makes these devices possible is the vast array of
‘central processing units’ (CPUs) now available,
chips designed to run an operating system and
mountains of software code.
But as the sliding fortunes of the PC collide
with rapidly growing demand for faster
artificial intelligence/machine learning, plus
the threat of high-tech ‘quantum computers’
waiting tantalisingly on the horizon, there’s
an alternative computing technology major tech
companies are taking seriously. It’s appearing[ A TECHLIFE TECH PRIMER ]THEY’VE POWERED COMPUTERS SINCE THE MID-1970S, BUT IS THE REIGN OF THE CPU COMING
TO AN END? TECHLIFE INVESTIGATES THE ‘POST-CPU’ TECHNOLOGY CHANGING HOW WE COMPUTE.
[ DARREN YATES ]in everything from retro-gaming consoles to
high-demand applications, such as search and
machine-learning and in one particular form,
comes with a feature that CPUs can’t match —
the ability to create your own chips.ALPHABET SOUP
Our modern lives are filled with digital
technology, virtually all of it run by small chips
or ‘integrated circuits’ that perform all manner
of functions, from starting your car to powering
your toothbrush. Those chips can be broadly
grouped into three categories that read like
an alphabet soup.
Application-specific integrated circuits
(ASICs) are chips that usually just have one job— it can be almost anything, but each chip
does just that one thing. It means they’re
exceptionally fast at what they do because the
required function is hard-coded into the chip’s
electronic switches or ‘transistors’. This
single-function operation also makes them
very power-efficient because only the transistors
required for the job are included, ideal for when
you’re using ASICs by the thousands in large
installations or on their own in battery-powered
devices. But the drawback of ASICs is that once
you’ve made the chip, it can’t do anything else.
Need another function? You need another chip.
At the other end of the scale are the
traditional microprocessor chips — the CPUs
that power your phone, PC, tablet, gamesGoogle’s Tensor Programming
Unit is aimed at accelerating
machine learning.