[ 076 ]
Famicom cartridges, but no longer relies on
‘new old-stock’ hardware. Again, it uses an
FPGA chip — the Altera Cyclone V — to mimic
the features of the original Nintendo hardware,
but goes further to support extras like 1080p
output over HDMI.
The NES Classic Mini might give you 30
games for $99 (if you can find one), but the
$US449 NT Mini’s compatibility reportedly
ensures it can handle over 2,000 NES-compatible game cartridges. More importantly,
it shows that FPGAs not only can replicate
long-retired hardware, they’re sufficiently
advanced to create custom CPUs.THE ACCELERATED BOOM
However, the spotlight shining on FPGA tech
over the last few years is less about retro-gaming
and more about the boom in machine learning,
AI and the ever-growing need for speed.FPGAs and ASICs are often called
‘accelerators’, in much the same way early
PC video cards were often called ‘graphics
accelerators’, and for much the same reasons.
While FPGAs can be made to emulate genuine
CPUs for gaming consoles, they’re also known
for their ability to work in parallel, plus their
very low delay or ‘latency’ times, the time it
takes for data sent over a network to be received
and processed.
One application where latencies have become
critical is financial markets. Share trading took
a major leap into the future in 1971 when the
US NASDAQ became the first computer-traded
share market. These days, computers aren’t just
making trades, they’re deciding what, when and
how much to trade in what’s known as
‘high-frequency trading’ (HFT). A key
requirement of HFT is speed, using machine
learning to understand market data in real time
to make high-speed buy/sell transactions.
FPGAs have been increasingly popular in HFT
since around 2010 and research into improving
FPGA performance in HFT continues today.PROJEC T C ATAPULT
That latency and parallelism capability has also
seen another major market player show serious
interest in FPGA. Back in 2011, Microsoft
launched Project Catapult (tinyurl.com/
h5kc3oq), a research project to look at
post-CPU technologies for cloud computing,
which included FPGAs, ASICs and graphics
processing units (GPUs).
We use GPUs at home to play games on PCs,
consoles and phones, but they’re increasingly
being used in business applications for their
speed and parallelism. For example, Nvidia’s
brand-new GeForce GTX 1080 Ti graphics card
boasts 3,584 processing cores designed toDiligent Arty is a US$ 99 DIY
development kit to create
your own FPGA design.You can make your hardware
using FPGA boards from
Adafruit.com.Intel’s 4004 processor,
the first CPU chip (Thomas
Nguyen, CC BY-SA 4.0).This Intel Atom chip comes with
built-in FPGA programmable block.FEATURERISE OF THE ACCELERATORS