Bitcoin, the cryptocurrency powered by a decentralized peer-to-peer network of computers, has been hot this season. With the exchange rate bobbing around US $100, those involved in creating new bitcoins—and upholding the network that makes them valuable—have become locked in an arms race of sorts, seeking new, powerful machines that will enrich them but that could also destabilize the nascent virtual money.
Bitcoins exist only as records on a virtual ledger that’s shared over a global network, each node of which must agree to changes in the accounting—payments or receipt of payments. Arriving at this consensus takes massive amounts of computing power.
You can use your bitcoins even if you’re not plugged into the network that runs the operation, but there is a strong incentive to help out. At an average of once every 10 minutes, the software spits out a handful of newly minted bitcoins to one computer, as a kind of lottery payment to the people (referred to as miners) who run it. The way you enter this lottery is by solving “hashes,” trivial functions that reformulate meaningful data sets into unique strings of letters and numbers. Each time a computer completes a hash, it’s as if it has filled out »
another lottery ticket, choosing the numbers and hoping they match up. Only a specific hash will be accepted, and the first computer to find the right one gets the prize.
However, there’s one more feature in Bitcoin that controls the profitability of mining. “It has everything to do with difficulty,” says Jeff Garzik, one of Bitcoin’s core developers. As the hashing power of all the combined computers increases, the network makes it less likely that a particular machine will compute the right answer. In the lottery example, this would be like making people choose more numbers on a ticket. And if the hashing power were to go back down, so too would the difficulty.
As the price of a bitcoin has gone up, more people have been building and buying dedicated machines they call mining rigs, which has driven the hash rate higher. Back in January, when miners were getting less than $20 for their bitcoins at the online exchanges, the hash rate was around 25 000 billion hashes per second. At the rate of participation on 2 May, the network computed 69 123 billion hashes per second.
Brock Tice, a miner in New Mexico, says it all comes down to economics and whether mining is worth the electricity bills. “I have my own rack of miners in the garage here. I had shut them down for a while...but then when the price spiked, I fired them back up,” he says.
It stands to reason that as long as miners are getting more “real” money for their bitcoins than they lost in electricity bills— last year the Bitcoin network was probably consuming 100 to 300 megawatt-hours per day—they will continue to mine. And what’s more, they will continue to upgrade their systems to ensure a bigger payout. These are the ingredients for a hardware arms race—one that really began the day Bitcoin was created but is now on the cusp of a rapid acceleration.
Early on, when a bitcoin cost a mere 25 cents to buy on an exchange and few people were participating, the difficulty level was so low that miners could run Bitcoin using only a laptop’s CPU and expect to make a handful of new coins every day. Many of the people who caught on early subsequently funneled their profits into faster machines rather than cashing out. Before long, this was the only way to keep up. As the overall hash rate of the network ratcheted up and it became infeasible to mine new bitcoins with a PC’s CPU, miners began to turn to GPU-enhanced computers. GPUs, although they are less flexible than CPUs, can execute certain tasks thousands of times as fast, speeding up the hash calculation. Eventually, people realized they could save on electricity costs by mining with field-programmable gate arrays (FPGAs). These integrated circuits contain logic blocks that can be reconfigured after manufacturing. When configured for the specific task of hash hacking, the FPGAs proved more energy efficient than GPU-enhanced systems.
But the ne plus ultra of customization is an application-specific integrated circuit (ASIC) that is purposely built to execute the hash operation. Last month, people began running the first ASIC-powered mining rigs, a development that could push the hash rate of the Bitcoin network to 50 times what it was last year while consuming half the electricity.
“ASIC is the technological progression of the endgame for Bitcoin,” says Yifu Guo, the founder of Avalon, a mining-rig manufacturer based in New York City and Shenzhen, China.
ASICs can be costly to make in the relatively low volume needed to supply Bitcoiners. In mid-2012, months before Guo got involved, Butterfly Labs, another rig manufacturer, began taking orders for a rig powered by ASICs made using a 65-nanometer manufacturing process. (That process was cutting edge in 2006.) A few months later, though, the company missed its shipping date, and people were disgruntled. Finally, Guo approached the Bitcoin community with a proposition: If he could get enough money from people up front, then he would begin making his own ASIC-powered mining rig. “If people wanted us to pursue it, we would pursue it, because we knew we had all the connections. We just didn’t have the money,” he says. “And then, lo and behold, people sent in their money almost immediately,” says Guo.
Relying on a less-expensive 120-nm process (cutting edge in 2002), Guo had the first chips made and systems shipped within five months. Garzik, who received one of the first, reported that upon arrival, his unit was computing 66 billion hashes per second. That’s more than twice the speed of the best FPGA-based rigs on the market. Within nine days of mining on the Avalon rig, Garzik claimed that he had made back the $1299 he had spent to buy it.
Quite suddenly, Guo has found himself in a position of great influence and responsibility. Theoretically, the computational power in the Bitcoin network must remain well distributed in order to stave off malicious attacks. If one entity were ever to control more than 50 percent of the hashing power, it would gain the ability to commit a double-spending attack, a manipulation of the ledger that uses the same bitcoin to pay for two or more transactions.
Guo stresses that his first concern is the health of the Bitcoin network. Had he gotten into the business for profit, he argues, his crew would have just refunded people’s money, kept the rigs, and started mining on their own. He claims that shipping the machines cost him $3 million in potential profits. In return, he has a clear conscience, he says, knowing that he is helping to spread hashing power evenly throughout the network.
“The whole reason why I wanted to make ASICs was to decentralize,” says Guo. “I sleep really well knowing that I kind of keep the network going right now. But at the same time, I don’t want to be in this vanguard position. It’s like being Batman.”
The arms race is not over. This is just the beginning of the ASIC era of Bitcoin. Three hundred Avalon rigs have hit the market, a second batch is on its way, and a third batch is in the works. As more miners start plugging them in, the hash rate will increase and push the difficulty level ever higher, spreading profits thinner.
And when it’s once again time to feed the need for higher hash rates, either Avalon or one of its competitors will be there with faster ASICs. “We’re going to keep making smaller-node-size chips for Bitcoin. And we’re going to repeat this process over and over until we reach the pinnacle of ASIC technology,” says Guo. What a bitcoin will be worth then is anybody’s guess.
This article was updated on 20 May 2013.
About the Author