the monster machines that mine Bitcoin

[Eugen Leitl]

(so I was wrong, ASICs have already made FPGAs obsolete)

http://gizmodo.com/5994446/digital-drills-the-monster-machines-that-mine-bitcoin?utm_source=feedly

Digital Drills: The Monster Machines that Mine Bitcoin

 Eric Limer

Bitcoin! It's everywhere right now. Its value is dropping, spiking, dropping
again. More and more new converts are hopping in, buying a few coins and
trying their hands at the market, looking to make a quick buck with a
profitable exchange. But all the while, there's an ever-dwindling army of
specialists working in the shadows, painstakingly extracting more and more
digital doubloons from the cryptographic static.

What started as a pastime any computer could dabble in has spurred the
evolution of truly ridiculous powerhouse Bitcoin mining rigs. Meet the
monster machines more powerful than everything you've ever owned, combined.
By a factor of thousands.

The Hacked-Together Past

It all started so simply. Just after Bitcoin's birth in 2009, and before its
current surge of popularity, mining could be done by a run-of-the-mill CPU in
just about any old computer. Back thenband still todaybBitcoins are mined by
using computing power to solve a complex cryptographic equation. If your
computer is lucky enough (or really, powerful enough) to get a right answer,
you'll create coins. So given enough time, a CPU can actually get some mining
done. And back in the early days, if you could get free power to run it from
your dorm room, library, or office, and the exchange rate was high enough (a
buck or two), you could stand to make some spare pocket change.

So how did we get from there to giant rigs that consume $150,000 of
electricity daily? It was inevitable, really. Especially after Bitcoins
actually became worth something.

The all-out race for power isn't just some sort of freak side-effect of
Bitcoin's surging popularity, though. It's actually built right into how
Bitcoins come into existence. The cryptographic hashes miners need to solve
to mine Bitcoins weren't easy to begin with, and they only get harder as
miners throw more and more collective computing power at them. Bitcoin
self-regulates its difficulty so that there's always just a trickle, one
block of coins mined every ten minutes. No matter what. So it's not about the
quantity of power you've got, but rather your proportion of the whole. You've
got to keep up with the Joneses.

An average CPU can run four instructions per clock cycle, and offer a few
thousand to a few million hashes per second. This was never exactly great,
but a few years ago it was enough for a low-rent mining operation. But then
miners discovered that they could use their GPU instead, which can 3,200
instructions per cycle. The indispensable Bitcoin Wiki describes the shift
like this:

One way to visualize it is a CPU works like a small group of very smart
people who can quickly do any task given to them. A GPU is a large group of
relatively dumb people who aren't individually very fast or smart, but who
can be trained to do repetitive tasks, and collectively can be more
productive just due to the sheer number of people.

And because mining is a repetitive task better suited to muscle than smarts,
miners suddenly found themselves with an 800-fold increase in hashing power
on their hands. Before long, the number of GPU mining rigsbcombined with
Bitcoin's regulatory difficultiesbmade massive amounts of power not just an
advantage, but table stakes.

So where do you turn when everyone's competing with the same horsepower?
Volume.



This is a multi-GPU mining rig, living in rented warehouse space with three
tons of AC for cooling. It was in its prime in late 2011.

Two GPUs are better than one, and two dozen are better than two. But when
miners started building up to such a macro scale, other concerns started to
creep in. Finding space. Keeping systems cool. But the thing that really
stood to cut into Bitcoin profits wasband still isbthe cost of electricity.
And this concern turned miners' affections to yet another, different tool of
the trade called FPGAs.

While CPUs and GPUs had been co-opted and tweaked for mining,
Field-Programmable Gate Arrays were the first devices with Bitcoin in mind
down to the circuitry. FPGAs are chips have the ability to be re-designed on
the spot, down to their very connections, so they can be good at just about
anything with the proper guidance. This isn't consumer tech. Instead, it's
generally used in supercomputers, data centers, MRI machines, PET scanners,
and the like. But the sort of flexible savant-level specialization is also
great for mining Bitcoins, once you can teach the chip how to do it.



An 8-core, pre-configured FPGA rig. It cost $2,000 at its introduction in
January 2013. By then it was already very late to the game, offering 1/60th
of the power of the Avalon ASIC which was released the following month for
less.

The jump from GPU rigs to FPGAs didn't bring on a power-increase of multiple
orders of magnitude, in the way the jump from CPU to GPU had. It did,
however, provide a modest power bump, and actually reduced the amount of
electricity necessary to get competitive hash-rates. But perhaps most
importantly, FPGAs servered as a final vaulting point into the nextband
finalbmassive revolution in Bitcoin machinery: truly specialized hardware.

The Made-To-Order Future

The first Application-Specific Integrated circuit Bitcoin miners burst onto
the scene in February 2012, bringing a new paradigm of power along with them.
Unlike FPGA systems, ASICs are not malleable; they are set in stone silicon.
And for that rigidity, they offer 100 times the hashing power of a FPGA setup
and actually use less power. No conceivable system could ever have Bitcoin
buried any deeper in its technological soul. ASIC systems have turned the
Bitcoin mining rig from something necessarily hacked-together and
jerry-rigged to something necessarily made-to-order; no miner can fab his own
chips in the garage.

Ultimately, ASIC devices are the last great innovation in Bitcoin mining;
once you've specialized down to the chipset, there's nowhere left to turn for
a 100-fold computing power increase. And in that way, we're seeing the
beginning of the end of the gold rush, just as Bitcoin fever reaches a
fever-pitch. As ASIC units continue to penetrate the mining market,
competitive advantage is going to come primarily from more power efficiency;
not by mining more Bitcoins, but by spending less electricity to mine the
ones you can. The land-grab is almost over. Now it's how efficiently you work
it.

The Ferocious Flagship

Meet the Avalon ASIC. Money cannot buy a better Bitcoin mining machine. It
was designed and built front to back with Bitcoin mining and Bitcoin mining
alone in mind. It cannot do anything else, but it cannot be beat at what it
does. Inside its unassuming grey case is an army of specialty chips that
promise 65 gigahashes per second. This is $6,800/`8?72.36 equipment (at the
time of this writing) that, properly utilized, stands to be worth much more.
It's the beginning of the end of revolutions in Bitcoin mining. It is the
endgame of an arms race.

As of now, Avalon ASICs have been the only systems to actually ship, though
they're exceedingly rare, numbering only in the hundreds. Still, this handful
of machines now accounts for such a large share of the collective mining
community's power that they've all but completely destroyed the viability of
anything that came before them.

And while Avalon started the revolution, others are not far behind. Companies
like ASICMiner, Butterfly Labs, and bASIC all offer similar systemsbsome that
boast even more powerbbut none of those have managed to ship. By the time
they do, if they do (many are accused of being cash-grab scams), it may
already be too late to justify their multi-thousand dollar costs. Granted,
many of these theoretical machines offer a little more power than the
current, existent Avalon machines, but you'd rather a real machine with less
power right now than a concept with more later.


The Butterfly Labs BitForce Mini Rig SC, a 1,500 GHz/s machine. It's about 30
times more powerful than then current Avalon ASICs, or would be if it ever
exists; its shipdate continues to be delayed. With every passing month, that
higher hashrate becomes less and less of an advantage, especially considering
the machine's unlisted but doubtless obscene pricetag.

Bitcoin has always favored early adopters, bold ones especially. Because
mining difficulty increases to compensate for increased power, all advantages
are temporary. But it doesn't increase dynamically. Instead it hops up once
every 14 days, so advantages can be leveraged. The first Avalon ASIC to be
put to use paid for itself in a mere nine days, but every additional unit
follows will have more and more ground to make up.

In the meantime, the price of Bitcoin continues to bounce around wildly, and
whether or not it's up at any given time is going to dictate whether running
these behemoths makes money or costs it. So long as it's a safe bet that
Bitcoin isn't losing its value forever though, these suckers will keep
chugging along.

It's not impossible that some unlikely future eventbwidespread lack of
interest, or the destruction of a huge share of existing mining rigsbcould
shake up the Bitcoin mining scene someday, giving older methods a chance at
renewed relevance. But for now the future of Bitcoin mining is locked up in
expensive, specialized, multi-thousand dollar boxes that sit in the corner
and hum. And whether that's "stable" or "boring" all depends on if you've got
one.