Hacker News new | past | comments | ask | show | jobs | submit login

Nobody on HN pointed out who is behind the increase... This is BitFury who just launched a 40 megawatt data center filled up with their new 16 nm chips which reportedly achieve approximately 0.06 joule per gigahash. They also use immersion cooling which gives them an insane PUE of 1.02. So the mining capacity of this DC alone is ~650 Phash/s! We saw an increase of ~200 Phash/s in the last 30 days, so presumably they are at only operating at 1/3rd of their capabilities so far:

40e6 (watt) / 0.06 (joule/gigahash) / 1e6 (petahash/gigahash) = ~650 petahash/second

BitFury's immersion cooling tech:


https://www.youtube.com/watch?v=uV7MDhqNyXE&t=0m42s (shows the fluid boiling - starts at 0m42s)


BitFury's 16nm chips:


So less than half of one datacentre entirely owned by one company is enough to commit a 51% attack. I get that they say they are going to work on distribution these chips to ensure they don't end up in that position but let's be honest they just destroyed any claim that the network is powerful enough to avoid malicious control by one miner now. They've not even raised that much money.

This DC cost $100 million. Not exactly pocket change. http://www.coindesk.com/bitfury-details-100-million-georgia-...

Plus, with BitFury online, the cost of a 51% attack just raised to $200 million.

The only reason it is (currently, for a short time) hypothetically possible to 51%-attack the network with a budget in the low hundreds of million of dollars is because most of the miners are not using such efficient 16 nm chips. As the market migrate to these last generation chips, expect the cost to increase to $1+ billion in the next year.

People seem to forget that it is NOT in BitFury's best interest to perform a 51% attack because, bitcoin's value will plummet if they do that.

>People seem to forget that it is NOT in BitFury's best interest to perform a 51% attack because, bitcoin's value will plummet if they do that.

If BitFury can turn $200 million into $400 million now rather than over a couple of years, why on earth would they care about the side-effect of killing off bitcoin?

For the avoidance of doubt, I think that BitFury's best interest is BitFury; as long as they make a decent chunk of money, the end result for bitcoin is - and should be - immaterial to them.

Yes but in practice you can't convert that much BTC into USD that quickly.

>Yes but in practice you can't convert that much BTC into USD that quickly.

Would they even need to convert BTC into USD? Even ignoring the obvious opportunities for shorting an entire economy?

I mean, I'm kinda curious to know what effects they could cause by e.g. choosing to mine zero-block transactions for a few days.

I thought miners already try to mine zero block transactions first, then add in fee transactions? apparently you can chew through the nonce really fast..

Not exactly. Some Chinese mining firms maintain a stratum connection to pools in the US so they can get just the new block header when a new block is found on the network. They then mine on that header until the full block is downloaded and validated by their local bitcoind (which can take 20 seconds, to several minutes in the worst case) This has been referred to as "SPV Mining".

If they find the next block during the time they are mining only against the block header, that block will contain no transactions. This doesn't happen very often, and there are some other issues that can arise from the practice (a multi-block fork happened once), but miners believe that it increases their profitability so (afaik) they continue to do it. More info here: http://bitcoin.stackexchange.com/questions/38437/what-is-spv...

But if that was a predictable consequence, they can't turn $200million into $400million now, because other would anticipate the future state of things and the price would plummet before a couple years. Not that I think that is at all what is going to happen.

>because other would anticipate the future state of things and the price would plummet before a couple years.

Why would the price plummeting matter to them if they've already made their money?

The only way they make money (cash out) is if trust remains in the bitcoin platform. By doing 51%, they tank the platform itself. No liquidity. Provable and transparent majority attack is MAD in the strictest sense.

>The only way they make money (cash out) is if trust remains in the bitcoin platform.

Not if their cash isn't - somewhat ironically - tied up in BTC.

Really, I'm sure between the two of us, we could imagine a thousand and one viable ways of making a tonne of cash very, very quickly that wouldn't involve transacting in BTC at any point!

Not sure I follow you here. How would they make money from controlling BTC other than by selling it? ...which is impossible if they break it, as argued above.

As the mining rewards keep lowering it's getting to the point here someone is going to do this for the lulz. As to cost you can sell all your coins before doing a 51% attack. Short the market and at some point it might even be proffitable to do so.

Yeah, but how many bitcoins is USD$ 100 million worth?

229922 BTC according to Google's exchange rates.

See: https://www.google.com/search?q=100M+usd+in+bitcoin

Or you could just buy bit fury

Or be the sovereign state whose jurisdiction that data center is in.

You couldn't buy BitFury for less than 200 or 300 million dollar. After all they expect this $100 million data center to be profitable so I am sure they value their company at least 2x or 3x this.

It's not pocket change… except for government organizations.

It is profoundly unlikely a state-level actor would bother.

You can cripple Bitcoin transaction clearing with a thousand dollars to DDOS the network with spam. If only state-level actors had a thousand bucks lying around ...

Malice isn't enough. The only special power of a 50% miner is to consistently resolve double-spend attempts in its favor. If this happened, it would surely lower the value of Bitcoin. This, an attacker must be not just malicious but also irrational enough to forfeit the $800,000 in Bitcoin that it creates daily by virtue of its 50% control of the network. (6 blocks/hour x 24 hours/day x 25 bitcoin/block x $450/bitcoin x 50%)

Not true. A malicious 50% miner can cancel any transaction it wants, not just fix doublespend. Assuming the transaction was sufficiently recent (the more hash power they have the more such a transaction can be in the past).

So they can "unspend" their own bitcoins, as follows : use 51% of your hash power to create a parallel blockchain, which is not published. Include all transactions, except the one you used to buy a TV. Put something else in it's place. Keep doing this until the TV is shipped/arrived. Then publish the parallel blockchain. Boom. Unspent.

Parent's point is that if BitFury did that, then nobody would trust Bitcoin anymore, thus crashing the price.

If BitFury resolved a single doublespend transaction differently from the "main" blockchain, people would(/should) realize that it now has the power to cancel transactions, and stop trusting bitcoin ASAP. At that point you can no longer trust the blockchain, since you can no longer know you're looking at the "true" blockchain, and not a fake one that has been presented to you to make you do something (e.g. pay someone real money).

In any reasonable person's version of "wait X blocks for confirmation" (currently mostly 3), X would be infinite.

>Parent's point is that if BitFury did that, then nobody would trust Bitcoin anymore, thus crashing the price.

If they make good money from it, why would BitFury care about bitcoin's long-term outlook?

they make money by mining bitcoin, if the value drops then their return, even if they sell all BTC immediately, would be affected. They just spent $100m on a new mining facility so to trash bitcoin would be to write off that investment. They cannot make good money by screwing over bitcoin without screwing themselves over.

Best case scenario is that at the end of the effective life of this DC BitFury decide to doublespend their coins to allow them to spend all bitcoins once, however by that time their relative % of the network will be less than 50%.

>They cannot make good money by screwing over bitcoin without screwing themselves over.

Why not? Surely this is more a lack of imagination on your part, rather than a hard fact?

Observation is that pretty much nothing crashes the price of Bitcoin.

* During the transaction spam DDOS attacks, the price went up even though it was literally unusable.

* The present price seems sustained by something that looks very Willybotish running between OKCoin and Huobi. https://www.reddit.com/r/Buttcoin/comments/3vnjgk/what_drive...

The remaining American Bitcoin traders are certainly gullible enough to keep buying and trading a 51%-compromised coin. (I mean, there are people who still think Paycoin could make a comeback.) But American traders are a sideshow - all the action is in China (miners, actual traders). So the question would be: will Chinese speculators keep gambling on a 51%-compromised coin?

(And of course the MMM ponzi buyers, whose judgement is sufficiently bad that they wouldn't even understand the problem.)

maybe people will trust bitcoin more if you put the right spin on it (bitcoin is now regulator-approved, etc.).

The attack you just described is semantically identical to a "double spend".

It's not necessarily 'irrational'. A 51% miner could profit from Bitcoin's downfall, or by threatening others, or even by bluffing an attack. See, e.g., the Goldfinger Attack in https://www.cs.princeton.edu/~kroll/papers/weis13_bitcoin.pd....

Stupid question: if control of 50% of the hash power is so unimportant, why does Bitcoin rely on hash power anyway? Why not simply have a consensus system without proof-of-anything?

If you remove proof-of-work, then the question then becomes: what else do you measure consensus by? A few other crypto currencies have tried other metrics but they all aim to for the same thing: removing the need for human trust.

Well, consensus between whom? How would you do distributed, anyone can participate at any time, byzantine consensus but still prevent Sybil attacks?

I think the reason is because then there would be no way for new coins to be mined and given to those who are rewarded for ensuring the security of the network.

>The only special power of a 50% miner is to consistently resolve double-spend attempts in its favor.

No, this post, and the handy table inside, explains the various attacks possible at different percentages of the hash rate:


Short first hope the exchanges don't fold and run it into the ground with double spends. Alternatively just attack it because you feel scared about it but I do t believe anyone in power is since it is not going anywhere

At any point since bitcoin's inception there have been plenty of major govts that could have easily afforded the resources to pull off a 51% attack.

A real bad thing here is the asymmetry. To prevent 51%, the network has to burn enormous amounts of energy 24x7 forever while the attacker only needs his energy for the duration of the attack.

Counterpoint: physical security largely works this way too.

Well, things work much better in crypto (not proof-of-work, but the "real" one).

And now a single company can. Progress!

It has always been the case that a single company could 51%-attack the network. No regression here. In fact the cost has continually increased over time. So, yes, progress.

Not less than half of their datacenter, but less than all of their datacenter. See the current hash rate here:


I will forever refuse to spend my 0.5 BitCoin just so one day I can say to my grandkids in a creaky old voice, "See! I mined that myself on my own computer in a couple of nights!"

"What are you even talking about, Bitcoin? And what's the big deal about the $0.5 you got there?" :)

How those 40 megawatts of energy consumption compare to known companies? How much energy for example consumes an average cloud storage company? Or a corporation? 40 megawatts sound a lot but it would be nice to have a reference.

After searching around, it looks about average for a large data center. This link[1] estimates Google's data centers require between 50-100MW.

50MW is enough power to supply ~14k homes from different references I've seen.

I'm curious about what normal data centers use for backup power since natural gas and diesel generators don't get much bigger than 2MW. Or maybe they just don't have backup generators.

[1] http://www.datacenterknowledge.com/google-data-center-faq-pa...

They do. Natural Gas fueled generators are practical at 50MW capacity. GE basically sells a 747 engine (CF6) driving a generator, known as the LMS6000

[0] https://en.wikipedia.org/wiki/General_Electric_LM6000

[1] http://www.geaviation.com/marine/engines/military/lm6000/

I understand that gas turbines can produce that kind of power, in fact I used to work on some pretty large turbines and turbo generators in the 500-1000MW range. However they aren't used as backup generators where they have to be online in a minute. These are the types of generators I was referring to, https://powergen.gepower.com/products/reciprocating-engines/...

More batteries to get over the slower generator startup time, and likely quite a few smaller generators to do the best they can to keep the batteries going if the big generator is slower than expected to startup that day.

That's probably the most awesome thing I've heard tonight.

That is cool. Wonder what the efficiency of one of those will be.

40%, it's in the linked Wikipedia article.

I ran a ~10MW capable facility for a while, we had 8 2MW Cummins gensets that would run in parallel. Extra units for redundancy as sometimes units would be offline for oil changes or other routine maintenance.

Maybe want to checkout engines from Wärtsilä.

They have gas capable engines (50DF series, 50cm cylinder bore) with ratings up to 975kW/cylinders and up to 18 cylinders or 17MW per genset at 60hz.

They're not the only player in that field. Fairbanks-Morse sells the Colt-Pielstick PC2.6 line with 750kW per cylinder at 600rpm and up to 16 cylinders for 11.5 MW @ 60hz. They also sell MAN 48/60 engines that go up to 21MW and MAN natural gas 51/60 G engines up to 18.5 MW.

40MW = 40,000KW Typical power density in a colocation facility these days is approximately 5kW per full cabinet (sold as 30A @ 208V or 60A @ 120V at 80% utilization). So, approximately 8000 full cabinets at average density, which would be enough power for one of the 10 largest known facilities in the world. Most likely though, the power density is going to be much higher than typical with bitcoin mining, and could be as few as 1333 actual cabinets at the high end of density of 30kW per cabinet.

Are those crazy jargon words meant to translate to $ ? If so, is it profitable? Sounds like they've just cut their lunch by 40% too.

Right at the moment the current hashrate is 723 Phash/s, so BitFury hashing at 200 Phash/s rate would get ~28% [1] of the newly generated bitcoin in average, i.e. 0.277 * 144 [block/day] * 25 [BTC/block] ~= 996 [BTC/day]. (Assumes that the difficulty approximately keeps up with the hashrate, which seems to be fine to me.) That translates to more than 400,000 USD per day if BitFury manages to sell them at the current price (highly unlikely) without disrupting the market (impossible at all), but yeah, that gives the rough order of magnitude.

[1] This also means that putting 650 Phash/s gigs into the network is not as profitable as it seems, since it will double (55%) the ratio while three times more expensive (if it linearly scales).

Since the amount of Bitcoins introduced into world daily is near constant (except for the scheduled halving of the rewards every few years), SOMEBODY is selling most of the freshly mined coins without disrupting the market.

It doesn't really matter how high the difficulty is or who the miner is the coins are mined and most of them sold.

There might be a small fraction of coins which are still mined in pools using outdated miners and those people don't bother selling them coins(fractions of coins that is).

It has been accepted wisdom that most mining operations sell most/all of their mined coins.

It's a race to the bottom. It will end with the party that manages to just win that race by making the smallest profit possible at the highest efficiency of Joules per hash computed. Bitcoin is interesting for many reasons, the real-world effects of a couple of configuration settings and some cleverly picked auto-scaling parameters are immense. The number of orders of magnitude that the protocol has survived with minor tweaks to date is very impressive.

Anywhere in the USA who has to pay USA power prices is going to be at a disadvantage by that alone then. It would seem that locating the hashing equipment to the place with the cheapest power possible would be a logical step when things start to get cut that fine. Unless that could be overcome with clever usage of naturally generated local power (such as solar power).

You'd think Canada would have more major data centers. They've got a lot of extremely cheap hydro power. You could surely lay ultra-high-speed fiber along the same right-of-ways that the transmission lines use. Also, it's cold as hell more often than not, so cooling is less of an issue than putting them out in the desert somewhere.

Bitcoin miner's dilemma: Should I mine where it's hot for the free solar energy or should I mine where it's cold for the extra refrigeration?

Photovoltaic power is actually less efficient where it's warmer. The ideal for photovoltaic generation is somewhere cold but sunny. And I'm not aware of too many solar thermal plants that are online and profitable versus other generating methods.

Cold air but hot ground. Take advantage of geothermal and hydroelectric in the arctic and near-arctic regions.

Iceland! Active volcano geothermal power and cold climate.

USA (Washington state) has the cheapest power in the world (at volume -- places like Iceland are cheaper/nearly free, but with too limited capacity for large scale mining).

Really limited capacity? They run aluminium smelters in Iceland, I can't imagine bitcoin mining requires more power than that.

At that price point, yes. Maybe the problem has been not enough power required... I don't know. I just know that most bitcoin companies in Iceland have moved or are moving to get cheaper power elsewhere.

Therein lies the problem: New users pay the average price, but new capacity has to be built at the marginal cost.

Washington and Oregon sell a lot to the SW. Google just built a big data center here near a dam from a huge city for power reasons.

There have already been articles about people doing that a couple years ago. I believe they were located in a cooler climate as well, made cooling the datacenter a bit cheaper so more power was going to computing than maintaining the computing equipment.

I'll look for that article now.

EDIT: Not the article I was looking for, but it covers it and was from 2013. This guy was using the cheap Icelandic power and the cold Icelandic air to his advantage.


> It will end with

What you describe is price-based competition. It’s probably the most documented situation in economics, and it present in many markets. It is a race to the bottom, but rarely ever ends; when it does, it is generally by disruption (something people on HackerNews know a lot about). In the mean time, it drives technical innovation after technical innovation—but I doubt, after 40 years of Moore’s law that it will suddenly end.

I'm wondering, isn't this going to put lots of current miners out of business, as their return-on-watt used to mine bitcoins is severely cut? Doesn't this risk to make BitFury even more dominant in hashing power, until somebody else can get comparable economies of scale?

Are these systems totally self contained? Seems a bit unsafe to have fire retardant chemicals vaporizing constantly in a confined space. Has anyone given any thought to the health effects of these type of data centers?

Could you translate some of that into an estimate of X coins mined per day and Y kwh used per day? Thanks.

The best way to think about the mining market is that the number of Bitcoins available to be mined each day is _constant_* no matter what the total hashing power deployed is. As a miner your share of that fixed pool of available coins is determined by your hashing power relative to the overall hashing power of all miners. So if you control 30% of the hashing power you should on average get 30% of the coins mined each day. Your profits are then determined by your costs and the value of a bitcoin when converted to the currency your costs are denominated in.

* The constant amount actually changes periodically when the coinbase reward is adjusted downward but that only happens about once every 4 years and historically exchange rate price appreciation has outstripped the reduced coinbase mining rewards. eventually the coinbase reward will go to zero and the number of coins that go to miners each day will be their relative share of the total fees being paid by users transacting on the network. Miners choose which transactions to include in a mined block so in the future a large miner may have some pricing power over transactions because they could refuse to process any transaction with a fee that falls below some threshold.

The question I haven't seen answered that I'm intrigued by: Assuming I put my 2015 pretty-normal-specs (i7 processor, 8gbs ram) laptop to work bitcoin mining 24 hours a day, how long before I have a single bitcoin?

Buy a scratch off lottery ticket you would have a better chance of coming out ahead.

At this point, basically never. Sad but true.

I find this style of commenting passive aggressive... "Nobody has pointed out this great fact?! I guess I will be forced to do it." Just lead with "This is BitFury..." we don't need to know how annoying it is for you to have to post this.

(Note: Since most HN readers are technical, this comment is being downvoted by bitcoin enthusiasts among them. You won't see many replies from them to this comment, since their reaciton is political and not a technical statement. Still, I'll keep it up. This comment is currently at -2.)

To put this into context, by comparing it with a centralized version, an equivalent number of actual transactions could be done at a bank with a crud app running on a $5 complete desktop PC which the Raspberry Pi foundation just released[1]. It is a 1 ghz computer with 512 MB of RAM. It draws up to 0.7 Watts.

Wait, wait, wait. did I say an equivalent number? Since Bitcoin is limited to 7 TPS (7 transactions per second), I should modify this to 100x more transactions (700 transactions per second), if you wrote it in C, or if not a hundred times, then at least ten times as many in python. If it were a crud app running at a bank, a $5 computer could do 100 times the transactions that 650 Phash/s of wasted effort collectively give. And 40 megawatts.

Bitcoin is a supreme waste of resources through the proof of work hack. It is like emulating a GPU in a CPU...in javascript. I mean, sure, you can do that. You can run Open GL by emulating a GPU in single-threaded Javascript.

But it's a stupid, wrong solution. The basic idea sucks. It's bad.

(by the by my target price for btc is $600,000 based on a comparison with how bad gold is physically, and the market cap of gold. :-P.)

I love fiat currencies, they're one of the great achievements of modern civilization. (This sentence isn't ironic, it's actually how I feel.) Despite my target price I don't hold any bitcoins at all. I hope some of the numbers I've given can put into perspective why.

[1] https://www.raspberrypi.org/blog/raspberry-pi-zero/

I believe that your being down-voted for writing:

> But it's a stupid, wrong solution. The basic idea sucks. It's bad.

instead of explaining that the cost is the cost of guaranteeing the authenticity of the transaction. HackerNews likes technical comments, and will happily burry expletives.

You could find more compelling example, such as estimate the ecological cost of having most of the current world-wide banking transactions on BitCoin, and prove your point.

You're wrong. Even without any mining, running a full node requires 2GB of RAM. Even if you do less than a full node, you'll find it hard to fully process 100x transactions on less than 2GB.

Also, you don't seem to understand the point of PoW.

You misunderstood him. He was talking about running a non-bitcoin program on that small computer.

He's talking about running something that does bitcoin, but without the inefficiencies of PoW. I pointed out that the RPi 0 doesn't have enough power to handle that scale, even if we took out the PoW stuff.

What the expense of mining buys you is censorship-resistance.

> Since most HN readers are technical, this comment is being downvoted by bitcoin enthusiasts among them.

No, you're being downvoted by technical people because you're missing the whole point of decentralized digital currencies. Educate yourself: https://en.wikipedia.org/wiki/Cryptocurrency#Timestamping

> I love fiat currencies, they're one of the great achievements of modern civilization.

Yeah, unless you want to export some Cuban cigars from Germany to Denmark: https://en.wikipedia.org/wiki/Society_for_Worldwide_Interban...

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact