As Bitcoin Infrastructure Booms, Mining Heads to the Data Center
January 21st, 2014 By: Rich Miller
This is the first of a two-part series on the boom in Bitcoin computing infrastructure, and what it means for the data center industry.
Emmanuel Abiodun once mined for Bitcoins on a desktop computer in his home. He’s now running 160 powerful computers in server space in Iceland, where the machines are cheap to power and cool. Later this year, Abiodun expects to have an empire of 4,000 bitcoin mining rigs spread across two continents, eventually filling nearly 5 megawatts of data center space.
“To be able to scale, you’re going to have to look at a data center,” said Abiodun, the founder and CEO of London-based CloudHashing, which leases server capacity to customers hoping to generate bitcoins. “We’re definitely seeing that trend.”
The journey of Abiodun and CloudHashing reflects the larger story of the Bitcoin network. After getting started in garages and server closets, bitcoin mining is moving into data centers and the cloud. Some traditional data center providers will benefit, but this transition also has the potential to enrich a new generation of entrepreneurs emerging from within the bitcoin community.
Over the past year, the computing power supporting the bitcoin network has soared, creating a powerful global network backed by 150,000 petaflops per second of computing power, roughly 600 times the combined power of the all the supercomputers in the Top500 list. Practitioners of Bitcoin mining – the term for using data-crunching computers to earn newly-issued virtually currency – are adopting more powerful hardware, pooling their efforts and seeking to slash their power bills.
An Opportunity for the Data Center Industry?
As this trend continues, “production capacity and operating efficiency will drive profit margins,” writes Jeff Schvey, a bitcoin analyst at The Genesis Block. “The natural evolution of this will be large data centers that can take advantage of economies of scale.”
The increasing industrialization of bitcoin infrastructure presents an opportunity for the data center industry. Large mining operations are beginning to follow the example of their forerunners in hyperscale computing, shifting compute capacity to remote areas with cheap power, including Iceland and central Washington.
But the bitcoin network may develop along several tracks, and not all of them involve traditional data centers. Some entrepreneurs in the bitcoin community are developing custom facilities to house high-density hardware, ranging from makeshift server farms in warehouses in the Pacific Northwest to futuristic racks of sleek, liquid-cooled immersion rigs in Hong Kong.
One thing is certain: Bitcoin infrastructure will need to become more energy efficiency. Recent estimates from Blockchain.info suggest that the global bitcoin network has been using between $12 million and $15 million of electricity each day, while generating $5 million to $6 million in new Bitcoins.
“Bitcoin mining is very energy intensive,” said Abiodun. “Power is key, because it affects your profitability.”
Bitcoin is sometime referred to as the “Internet of money” – a platform using cryptography and software to offer an alternative currency and payment-tracking system. At its heart is a huge distributed computing network that verifies each transaction. Participants in this online ledger – which includes individuals, corporations and mining collectives – are rewarded with new bitcoins, which are issued about every eight minutes. This ability to discover new coins has prompted the “mining” analogies, which some say are misleading.
“I hate the term,” said Josh Zerlan, Chief Operating Officer for Butterfly Labs, a Bitcoin hardware specialist. “We’re not mining. This is transaction processing.”
Cloud Hashing: Building Big
CloudHashing has been the first major bitcoin specialist to go public with its use of data centers. The UK-based company has set up shop near Reykjavik, Iceland in a data center operated by Verne Global. The company’s presence in Iceland was profiled in The New York Times, and is about to get a lot bigger.
“Our mission is to bring Bitcoin mining to everyone in an easy and affordable way,” said Abiodun, who said the service has grown from 800 customers in September to 5,000 in early January. To meet that surge in demand, he has ordered 1,500 custom mining rigs using ASICs (Application Specific Integrated Circuits) to crunch data for creating and tracking bitcoins, which he says will boost CloudHashing’s Iceland operation to 2.5 megawatts of power capacity.
JeffPosted January 21st, 2014
The big downside to mining BTC is that the amount of BTC one processor can mine (known as difficulty) is effectively divided equally among all the processors participating (assuming each processor can hash at the same speed). So if for example you have 10k processors and your competitor has 10k, and they upgrade to 20k, they will then get 2/3 the mining revenue (the revenue total has not gone up, the share just shifted) unless you add another 10k just to keep up.
The real opportunity now is figuring out what to do with all those ASIC devices once the value of BTC pops…
What is the useful lifespan of hardware? What is the term for a lease?
Is density factored into roadmap on hardware and product planning with the data centers?
The way the article is written it is pretty clear that the hardware is the driver of the business. Master the hardware design and manufacturing and deployment and you have a leg up on competitors. That said, what is the useful life on the chips and the design? 2 years tops? One of the companies flamed out after a few months because they appear to have been stymied by the rapidity of hardware designs. Short lifespan/short useful runs makes for volatility the underlying real estate business.
From the data center perspective – the challenge will be looking at lease terms and density. You have new companies with rapidly changing hardware requirements that may obsolesce a facility before the end of term.
If hardware changes that quickly, then the hardware in the first 10 cabinets of a row will likely be different densities than the next ten and have different cooling requirements, and if you outpace the ability of the facility to (1) keep up with densities and (2) keep up with the flexibility required for cooling and or constantly dealing with hotter hot spots then you may have a situation that will be hard to fix.
I could also see a bait and switch (for lack of a better term) on signing up for 3MW and burning through that in 1-2 hardware refreshes in two years and no expansion because of the heat footprint increases.
I hope the bitcoin miners look at containers as options – fixed number of racks, fixed cost, onboard cooling that can be dialed up from 250Kw-750 Kw in the same form factor and down the road the companies can order 2×250 Kw’s to get rolling, roll in another 250, and then add a chunk of 500Kw containers as they designs change. They may be one of the few options that can service the need that quickly. When it’s time to move up again, they recycle the container, or create a secondary market, or retool them along with their hardware.
Integrated bitcoin hardware and data center containers for speed of manufacture & maximum flexibility across densities and commonality of platform would be safe way to go. Then just lease power without the facility overhead, it could be a very compelling proposition. I’m game to model it out…
MikePosted January 21st, 2014
Mark this might help you, Asicminer are working on containers see https://bitcointalk.org/index.php?topic=99497.msg3516608#msg3516608
PepePosted January 22nd, 2014
If you really want to take mining to everyone’s hand you can develop more user friendly mining software, or cheaper, afordable and reliable ASIC…
when you buy equipment for mining, is always very close to break even or loss, and always is below to buy and hold bitcoins. But in the long run at some point you’ll get the same in btc. But in a time limited contract you can calculate more o r less how much BTC you’ll get from mining and compare with buy and hold the BTC in the same time.
And contracts for the prices that Cloudhashing offers, the customer will get about a 20% of the BTC compared to buy and hold… so the only profit is made be cloudhashing as they keep the equipment mining and renting in.
With the renting profit the get the mining product in advance so they are mining before the network…
So dont trust this people saying they do it for the customer, all that is a lie the customer is at the end paying to maintain cloudhashing’s mining farm. And that is the business model…
Mark: A few weeks back we wrote in more depth about the ASIC Miner Hong Kong project, which uses immersion cooling but deploys it in rack-mounted tanks. Check out the photo here:
As Mike noted, they’ve also looked a container model, which makes great sense for this use case, especially for those wanting to use traditional cooling rather than immersion.
Lease lengths and bitcoin are an interesting topics, due to the built-in changes in the difficulty and block rewards over time. Watch the site for more on that, probably tomorrow.
Miners have been exchanging a lot of their new Crypto coins, for gold,silver,platinum, and palladium at sites such as mintagemastermind.com
Jeff: Re the hardware, I’d be interested in hearing from folks with specific experience with Bitcoin ASICs and whether they can be repurposed. There’s been chatter in online forums about efforts to adapt ASICs for mining Litecoin and other Scrypt-based cryptocurrencies (which were designed to be accessible to miners using GPUs).
JeffPosted January 22nd, 2014
One of the more interesting things I’ve read is that since the ASICs are good at hashing they would be great at cracking encryption keys otherwise thought to be impossible to brute force in any timespan shorter than the predicted life of the sun (outside of the creation of a true quantum computer). One particularly paranoid write-up even postulated that Bitcoin is already a front for a [insert favorite covert agency here] plan to create a rainbow table of hash reverses. Bitcoin’s inevitable downfall could be a pretty devastating thing to happen to secure communication, if any of that is true. I suspect if it is, an arms race will ensue with servers of any value needing onboard ASICs for elaborate encryption, just to stay far enough ahead of the brute decryption capabilities that are out there.
Last I checked, the Bitcoin mining pool (not including any alternative cryptocurrencies like you mentioned) is hashing 15 quadrillion times a second (thats 15,000,000,000,000,000,000). Nervous yet?
“One thing is certain: Bitcoin infrastructure will need to become more energy efficiency. Recent estimates from Blockchain.info suggest that the global bitcoin network has been using between $12 million and $15 million of electricity each day, while generating $5 million to $6 million in new Bitcoins.”
Energy efficiency will not solve the overall energy consumption rate. When you increase the efficiency of the overall network, then it lowers the entry barrier for new miners and rigs to compete on price. This hypercompetition results in favoring those miners who can operate on the thinnest profit margins. This translates into mining profits being just slightly above mining expenses, resulting in the *same amount* of power consumption per coin mined. As the price of each bitcoin increases so does the incentive to spend more on electricity. Assuming a bitcoin price of $100,000, and a mining rate of 100 coins/hour (it’s currently 185 coins/hour), and a global average rate of 0.10 Kwh (1000 watt-hours), translates into 50 Gigwatts of continuous power, or 1.25% of global electrical power output! At a $1 million per coin that works out to 12.5% of global power output! Clearly, these power consumption rates will in turn act as a constraint on bitcoin’s ability to scale to handle demand, thus limiting it’s future price.
As a bitcoin investor I am very troubled by this “energy problem” and I’m amazed no one in the community has addressed it. Again, increase energy efficiency of mining rigs does not lower the cost per coin, it only increases competition for that coin, and thus the energy required to mine it.
Paul, wasn’t that $15 million number completely flawed? I believe it was just in some statistic on blockchaininfo.com that took the hashing power multiplied by some random GPU’s power consumption, for fun I assume. It has been taking down in the meantime. Whoever broke that story was probably only after creating some headlines.
There have been arguments over the true cost of Bitcoin mining (and hence production of new bitcoins) versus existing infrastructure of banking data centers and the cost of printing, dispensing and exchanging physical money all around the world. One popular but oversimplified approach to estimate total electricity cost is to multiply the estimated total network hash rate by the electricity required by popular GPUs for each mega hash per second (MH/s). However flawed this method is, due to the ever changing mining market and devices in use and different electricity cost throughout the world, none of these estimations take cooling and infrastructure electricity consumption into account. While it is not unheard of that hobby miners reuse some of the heat to keep their homes warm in colder climates, it is safe to assume that large scale mining is carried out in either commercial or purpose built data centers and not very different from commercial data centers in this respect – very little of its generated heat is reused for a purpose.
Actually, it was our story that mentioned the $12 million to $15 million number, which indeed came from blockchain.info (which Paul cited in his comment). I realize that’s just a guesstimate, and the 15 cents per kWh rate seemed pretty high to me, but may reflect rates on miners outside the U.S.
AlexPosted January 23rd, 2014
The $15 million was on Forbes too a few weeks ago, so I didn’t mean to address you with “creating headlines”. Not sure who broke it though.
What I am really asking myself is where all these 2kW 4U boxes are supposed to go? Is there even broadly available infrastructure in the US that can support this kind of density?
Good discussion, thanks for the input and reference articles esteemed colleagues.
I was in meetings in Chicago the past couple of days and the discussions were focused on a number of things – very little of which was solution or application focused. What I mean is that many data center operators think of what the data center is from a capital, construction, or ‘market’ based perspective. It’s still led by raised floor with the variables of density and size and the caveat of *we can build anything with enough time and the right financial model to back it.
Seems to me there is a vertical focus emerging that is high risk/high reward in the mining outfits. The gear is volatile and obsolete well before the end of a data center lease. That much volatility in a data center won’t play well from a leasing or power management perspective because the tenants are chasing the dragon for more power while the operator will likely run out of the power and cooling to service them AND the other less dense customers. It happened in a facility a former employer had in Minnesota – cages every 100 feet with small clusters of hotter and hotter gear. And that was years ago.
For someone that sees this for what it is – a great opportunity to work through these challenges and come up with a model that embraces the changes, the risk, can handle the responsiveness required to serve the dynamic nature of the clients – there is a specific application to be mastered and modeled here. I for one want to be a part of the solution.
I know the focus has been on the liquid cooled stuff and I get the benefits, but I have yet to see techs jumping up and down at the opportunity to wear aprons and boots to keep mineral oil off their clothes to swap out a blade. The additional infrastructure required in tubing, pumps, and enclosures adds points of failure from what I have seen and piping is expensive. In one project it made a $20M difference in price from piping alone. Maybe I am wrong, but I haven’t modeled it out vs what else I know and have modeled and I’m willing to learn.
Great stuff, and I would love to see more of these specific applications covered as it relates to data centers. Watts per square foot is not what it used to be…
Mining doesn’t have to be so power hungry. That’s what happens if you choose a proof-of-work scheme that’s computationally intensive. A scheme that minimizes computation and maximizes random access to GBs of memory, like Cuckoo Cycle (https://github.com/tromp/cuckoo), will require large investments in RAM but relatively little power.