Washington - Self-driving
cars. They're the future of transportation - and they're getting smarter all
the time. Thanks to advances in software and artificial intelligence, these
machines are now able to distinguish between cars and cyclists, or between
pedestrians and your pet.
Many can now "see"
just like you can, picking out objects and obstacles approaching ahead. All
that tech could eventually save lives, helping to prevent the 95 percent of car
accidents that safety regulators estimate are caused by human error each year.
But none of this would be
possible without a piece of hardware many of us take for granted in our own
home computers. It's a technology that traces back to the earliest days of
modern personal computing, one that people tend to associate more with
"World of Warcraft" than newfangled widgets on wheels.
We're talking about the
graphics processor. In mainstream PCs, the graphics processor - often found on
a graphics card - is what allows computers to draw all those pixels and
polygons that make up today's photorealistic video games. But as these
processors have grown ever more powerful, engineers have discovered their
utility in all sorts of nongaming applications. Graphics processing units - or
GPUs - have transcended their origins to become entire computers in their own
right.
"[The GPU] is now
powering everything from games to the visual effects you see in Hollywood
films," said Danny Shapiro, the senior director of automotive at Nvidia, which
accounts for roughly 75 percent of the $7.8 billion market for GPUs. GPUs, said
Shapiro, are central to "professional graphics, for automakers that are
designing cars, to doctors and researchers that are searching for cures for
cancer and using medical imaging techniques."
It's a sign of how big the
GPU business has grown that some 200 other companies work with Nvidia's
automotive unit alone. GPUs are even part of the brains behind artificial
intelligence, appearing in technologies like the Amazon Echo, which converts
natural human speech into data that machines can understand.
"The combination of
GPUs and a CPU are now available that can accelerate analytics, deep learning,
high-performance computing, and scientific simulations," Chris Niven,
research director for oil and gas issues at the research firm IDC, told ZDNet
last month. To understand why GPUs have
become so prevalent in next-generation technologies, we have to talk about how
they work.
Traditionally, the brain in
most PCs has been the CPU, or the central processing unit. These chips are made
by companies such as Intel. Apple has also been making its own, proprietary
chips for the iPad and iPhone.
The distinguishing feature
of this technology is that it's designed to run calculations serially, one
after another, very quickly. The rise of dual- and quad-core CPUs have expanded
their capabilities, allowing for more computations to occur simultaneously.
These chips are still ideal
for machines that only need to run a few processes at the same time. But when
it comes to technology like self-driving cars, where the computers are
constantly receiving and digesting information, multitasking becomes that much
more important. And that's where GPUs excel.
Computer researchers began
to discover the potential behind GPUs as far back as the late 1990s, when the
market was awash with dozens of competing chip makers. Their products found
their way into desktop PCs and gaming consoles like the Sega Dreamcast and
Xbox, enabling consumers to experience groundbreaking titles like
"Half-Life," "Quake" and "Halo." By
simultaneously and efficiently controlling the generation of shapes on a
screen, GPUs helped bring first vector graphics, and then individual pixels, to
life.
By the early 2000s, GPUs
were being pit directly against CPUs in computing tests, with some results
showing enormous promise for graphics processors.
"Researchers at
universities realized that, 'Hey, here is this low-cost processor that we can
apply to scientific and mathematical applications and get some acceleration for
cheap,'" said Jon Peddie, president of Jon Peddie Research, an industry
analysis firm.
One paper in 2002 found that
compared to CPUs, "the graphics hardware allows us to establish a
high-speed custom data processing pipeline. Once the pipeline is set up, data
can be streamed through with devastating efficiency."
The best GPUs on the market
today come with as many as 5,000 cores, said Peddie, not just two or four or
eight as with CPUs. While CPUs can process smaller amounts of information very
quickly, the advantage of GPUs has to do with scale - processing lots of
information at the same time.
Read also: Real drivers for Google's autonomous cars
This is why self-driving
cars find GPUs so useful. Through the use of optical cameras, laser and radar
sensors, cars look at their surroundings by taking many measurements per
second.
"It's 30 pictures every
second," Shapiro said. "Each picture, a single frame, is made up of
pixels. Each of these pixels or dots is a numerical value that says, 'What is
the color of the light there?' It's just a bunch of numbers."
GPUs like the ones found in
self-driving cars are designed to crunch those numbers and figure out that some
of those pixels represent an obstacle, whereas other pixels are lane markings
and still others are traffic lights. While GPUs weren't originally invented for
those purposes, car engineers began taking advantage of the technology's
parallel computing powers about six or seven years ago, according to Peddie.
"The original use of
GPUs in an automobile was for the instrument panel in the entertainment
system," he said. "It's only been recently that people have been
saying, 'Hey, we can do this, or that!'"
As GPUs become even more
powerful and gain even more features, you can expect them to crop up in even
more places. Within automobiles alone, many stand-alone processors that used to
handle just one function - such as the anti-lock brakes or the power windows -
will all someday be routed through a single processor, the GPU, said Shapiro.
And we'll see cars work increasingly like Tesla's automobiles, where you might
customize your vehicle by picking and choosing different software packages to
suit your driving style.
"You can almost have
in-app purchases to add new features that weren't there when you bought
it," he said. For gamers who've grown accustomed to buying expansion packs
to their software - also known as downloadable content, or DLC - this idea
might sound very familiar.