Bitcoin and the ASIC Conundrum

[Korbman]

NEW EDIT: Updated my paper to reflect data gathered on January 10th, 2013. You can read my new January Analysis here. The "TL;DR" version is that I increased my minimum from 217TH/s to 260TH/s, with a 12 month cap between 400-500TH/s.


OLD EDIT: Updated my paper to reflect all the new information that's come to light within the past 2 months. As a result, you can read my new October Analysis here. The "TL;DR" version is that I increased my minimum from 147TH/s to 217TH/s, with a 12 month cap between 350-400TH/s.


Over the past day or so, I've been digging into ASIC and what it might mean for Bitcoin and the community. Presented below is a small essay I wrote that sort of sums up my ideas...it was helpful to put words on paper to get a visual aspect on the situation.

Let me know what you guys think, what you'd like to add or change, and anything else. Discuss it! I'm happy to see counterpoints!




As many people know, Butterfly Labs Inc. has been producing custom field-programmable gate arrays (FPGA) specifically designed for the mining of Bitcoins.

The time is coming, however, when they are to be replaced by application-specific integrated circuits (ASIC), which is largely considered to be not only the next iteration in mining technology, but also the final frontier in hashing power before going Quantum.

At first glance, the idea is astounding; eye-opening, to say the least. Mining first began on the CPU, barely breaking into 1 Megahash (1 million hashes) per second (MH/s). Then came the introduction of the GPU and the surprising power that it holds, making Gigahashes per second (GH/s) a reality instead of a dream. And in recent months we’ve seen the push into FPGAs, thrusting us from 1-2GH/s into ranges 5 or more times that.

And we seem surprised that the next transition is to the upper echelons of GH/s and even the breakthrough to Terahashes per second (TH/s). That’s one TRILLION SHA-256 hashes per second. For a year, it seemed only large mining pools had the capability to break this barrier. Now we’re presented with the news that we can own this power on our own.

But is this too much power? How will the Bitcoin network react to such a drastic increase in speed? How will users react? What about the Exchanges and secondary markets?

It’s extremely hard to answer these questions, and as a result all we can do is speculate. Looking toward the past won’t be much of an indicator given the incredible volatility. And we all know looking to the future is nothing more than a best guess.

With a little digging, however, we can unearth some of these best guesses and see where it leads us.


Mining Pools and Their Top Hashers:
At the time of writing, the network hash rate sits at about 20 TH/s (varying by ±2TH/s every so often) and difficulty is 2190866 (set to increase 10.3% to 2416087 soon).

How are we managing to pull off 20TH/s? The top 5 mining pools account for approximately 60% of all hashing power. These pools are:
DeepBit at around 3.7TH/s
EclipseMC -- 2.2TH/s
50BTC -- 2.1TH/s
BTCGuild -- 2.1TH/s
Slush -- 1.5TH/s

NOTE: These rankings change all the time as workers come and go, stop and start. This information was obtained from: http://bitcoincharts.com/bitcoin/

I went through each of these pools trying to get an idea of who contributed the most. I wanted to see if I could figure out (or at least make an educational guess) what percent of miners were really behind the network hash rates we keep seeing fluctuate. To meet the ‘top tier’ requirements in my mind, the miner must be going at a minimum of 4GH/s (though I settle for 3.5GH/s).

DeepBit
There are roughly 7,200 (±500) workers associated with the various teams at DeepBit. 5,800 of these workers were situated in the top 45 teams (out of 750 or so), with the top team having over 1,200 workers alone.
When tallied up, these 45 teams (6% of the total) contributed 1.6TH/s out of the total 3.7TH/s (43%). In each team, roughly the top 3% controlled 70% (±5%) of all hashing power for that team.
TL;DR: 5,800 workers in the top tier of the pool, and 174 of them control 1.12TH/s of power (30% of total pool hash rate).

EclipseMC
Around 1,500 workers seemed to stay constant while I did my research. Out of these workers, the top 40 hashers made up 858GH/s. A nice 39% of the total pool’s rate of 2.2TH/s, yet only 2.6% of all 1,500 workers.

50BTC
A bit more difficult to determine total workers, but my calculations put it around 2,700 (±200). The top 20 (!!) hold 829GH/s of power, compared to the total pool rate of 2.1TH/s. In this case, we’re looking at 0.7% of workers contributing 40% of the pool’s power.

BTCGuild
This pool was nearly impossible to find statistics for. Given their total hash rate of 2.1TH/s, I’m giving my best guess that they have around 2000-2500 workers, so I’ll say 2250 in this case (though please correct me if anybody has a more accurate number). Their top 50 miners (2.2% of total) still manage to contribute 709GH/s, or 34% of total power.

Slush
Lastly we find ourselves with the immensely variable BitcoinCZ Mining (also known as Slush), which I’ve seen go around 1.2 to 1.7TH/s throughout today. For easier calculations, we’re going to assume 1.5TH/s is relatively accurate.
Although they have a reported 7,700 workers, they don’t have any solid information on top contributing workers or such statistical data.
However, we can average the other 4 pools “top tier” percentages, which comes out to 39%. Meaning, the top 3% of miners (230 here) contribute 585GH/s (or 39% of the total 1.5TH/s).


When condensed, we find that the top 5 mining pools contribute 11.6TH/s (though let’s round to 12TH/s) out of the total network rate of 20TH/s. Out of this 12TH/s, the top 3% of miners work to an astounding rate of 4.5TH/s, nearly 38% of all power.

DON’T FORGET -- These numbers are purely educational guesses based off of data I gathered over the course of 12 hours (or so) on August 20th, 2012. This WILL change over time...probably will be obsolete by next week


The Network
The thinking goes, if you can get an idea for the top miners, you can get an idea of who will be controlling the network when ASIC mining hits full throttle.

As I noted before, I only touched on the top 5 pools, which was roughly 60% of the total network. So let’s work out some math:
Added together, the top pools are supporting approximately 21,350 workers. At 60%, this would mean the total network is working with around 35,500 workers (I suppose ±2000 at this point).

With the averages above, we’re looking at the top 3% of miners to see how many actually contribute in any substantial manner. At 3%...
DeepBit contributes 216 miners
EclipseMC -- 40
50BTC -- 80
BTCGuild -- 70
Slush -- 230
...for a nice total of 636 ‘top tier’ workers. Added up, these 636 contribute a solid 23% of all network hashes.

If you look toward the network as a whole, the top 3% comes out to about 1065 (out of the total 35500). The theory is...if you’ve invested this much time, effort, and money into achieving the status of “97th percentile”, then the chances are that making the switch to ASIC devices won’t require much thought.


Speculation and My Incoherent Ramblings
Welcome to the final section of my research, where even educational guesses can’t predict the mercuriality of Bitcoin and the future of the network.

Yet I still feel compelled to try because the biggest question is still on my mind; Will Bitcoin still be profitable?
Yes! I mean, no. Maybe? Hopefully?

Let’s hash it out and see where we end up. Let’s assume my previous math was correct, my numbers are solid, and the percentage of top miners was accurate.

When the transition to ASIC devices is fully underway, I entirely expect to see a substantial drop in casual mining (workers going at around 300MH/s and under), which essentially accounts for about 60% of total workers (no math here, just pulled that number out of my ass to suit my idea of “best guess”).
If the top 3% (1,065 workers) are equipped with substantial GPU and FPGA setups, I can only presume they’re going to be the ones purchasing BFL SC Singles and “Mini-Rigs”. This leaves 34,435 “casual miners” with standard GPUs.

But not all 1,065 workers will be making the BIG purchase to get a Rig. My thoughts for that reside within the Top 5% of this tier, or about 54 workers. So here’s how it looks to me so far:
Normal   34,435
Singles   1,011
Rigs          54


When the transition is made and underway, invested users will stay by purchasing BFL SC ‘Jalapeno’ devices (for the most part, GPU mining is now obsolete). Upon losing 60% of normal workers because they can’t compete with high GH/s and TH/s, we end up looking like:
Normal   13,774
Singles   1,011
Rigs          54


This would mean a vast increase in hashing power and difficulty.
Jalapenos   13,774   3.5GH/s   48,209GH/s   
Singles       1,011    40GH/s    40,440GH/s   
Rigs           54       1000GH/s  54,000GH/s
                                         142.65TH/s Tot.


That’s right, 142.65TH/s. A 7-fold increase in hashing power, accompanied by an equal increase in difficulty to approximately 15,336,055.

And to top it off, we see the reward split to 25BTC per block coming up.

1. Buy Rigs, 2. , 3. Profit
Each BFL SC “Mini-Rig” goes for a flabbergasting $30,274 ($29,899 + $348 Shipping).
With the new difficulty and reward split, the Bitcoin Mining Calculator shows an income of about 1,000 bitcoins per month. And I like to assume that Bitcoins are worth something around $5 (since that was when it was stable for the longest, though I’m willing to entertain the idea that the price will double to $10 during the reward split).

If the Rig pulls down 1,500 watts, you’re looking at a usage of around 1,080KwH. In my area, at $0.147 per KwH this would be more or less $160 a month.

So $5,000 - $160 = $4,840 per month profit. Also known as a 6.25 month payback period, which is an extraordinary time (in a good way, I was expecting longer when I did my calculations).

Feasible? Seems that way. Practical? Not too shabby if you’ve got the money. Profitable? Eventually.

But if anything...nothing will change. The top miners will stay in their upper percentile as the majority of us fall to the wayside or buy up singles and Jalapenos to make up for our now lacking GPUs. It completes the cycle from MH/s to GH/s and moves us further into the world of Bitcoins. There will be turbulence to begin with of course, but when the market and network stabilize I imagine nothing will appear different.

Then again, I suppose that’s just wishful thinking



Here's a link to my original paper:
https://docs.google.com/document/d/1L8hKZinune9Cc7tWY9Pmx8J17BOxor7QNQAV03xROOc/edit

[Raize]

While the price will/can/could likely increase due to decreased availability but constant demand, too sharp of an increase will/can/could have a profoundly bad impact on mining as a whole as miners make purchases based on a ROI that is irrational. I tend to think there are a reasonable amount of people that don't mine and have purchased quite a bit of Bitcoin that they intend to sell once ASIC comes out (buy on hype, release on news) and then become miners themselves.

I also tend to think there's another ASIC player waiting for BFL to complete/finish and start selling their products so the competitor can immediately jump in and undercut their prices. And I think this competitor may have been ready to do this for the last year or so.

Still, you've got a lot of good points made, but you may want to account for some of the statistical analysis that has already been done based on BFL ASIC order numbers putting the increase at closer to 300 TH/s.

[_mr_e]

My question is... How easy will it be for someone to attempt a 51% attack using these? As difficulty goes up, the reward rate and therefore number of miners goes down, won't this be a serious problem?

[Raize]

My question is... How easy will it be for someone to attempt a 51% attack using these? As difficulty goes up, the reward rate and therefore number of miners goes down, won't this be a serious problem?

It is very profitable to mine right now (ROI of ~year or less). After the block reward halving, it will still be profitable, even at present prices. The only places where it won't be will be expensive electricity locations in Europe with GPUs.

[BCMan]

My question is... How easy will it be for someone to attempt a 51% attack using these? As difficulty goes up, the reward rate and therefore number of miners goes down, won't this be a serious problem?

It is very profitable to mine right now (ROI of ~year or less).

Sure, very profitable with YEAR roi.
 

[Korbman]

While the price will/can/could likely increase due to decreased availability but constant demand, too sharp of an increase will/can/could have a profoundly bad impact on mining as a whole as miners make purchases based on a ROI that is irrational. I tend to think there are a reasonable amount of people that don't mine and have purchased quite a bit of Bitcoin that they intend to sell once ASIC comes out (buy on hype, release on news) and then become miners themselves.

I also tend to think there's another ASIC player waiting for BFL to complete/finish and start selling their products so the competitor can immediately jump in and undercut their prices. And I think this competitor may have been ready to do this for the last year or so.

That's actually a very good point. I imagine that there are loads of people placing bets on ASIC in the Exchanges, who will probably join the mining game once they make it big.

And, like you, I suspect BFL won't have a monopoly over quality ASIC devices for mining...but I couldn't really find any thorough information to make a good guess about the competition or pricing, so I figured I'd just work with the numbers I knew.

Still, you've got a lot of good points made, but you may want to account for some of the statistical analysis that has already been done based on BFL ASIC order numbers putting the increase at closer to 300 TH/s.

See that's the thing, I was thinking about this while writing everything up yesterday. The way I thought about it, the big miners out there are already in the midst of making their purchases / trade-ins for the Rigs. Last that was posted, only 19 Rigs have been purchased out of my "54" predicted so far. The same thought goes toward Singles.

Let's call 142TH/s "low balling" it for now. I'm not sure it'll hit 300 right away (maybe over 12 months or so), but 200 is definitely within reason.
And at 200TH/s, mining is still profitable, but takes about 8 months to break even instead of 6.25...still not too shabby if you've got the financial backing and time.

[matthewh3]

Release of the first generation ASIC's isn't the endgame for the network hash-rate and difficulty as according to Moore's Law ASIC speeds should double every eighteen months.

[Korbman]

Release of the first generation ASIC's isn't the endgame for the network hash-rate and difficulty as according to Moore's Law ASIC speeds should double every eighteen months.

I didn't specify a generation

[freeAgent]

I am also worried about the massive increase in power bringing pools down.  Eclipse, which I'm a member of, is already having load issues.  If hashing power goes up exponentially, it will kill the pool.

[Korbman]

I am also worried about the massive increase in power bringing pools down.  Eclipse, which I'm a member of, is already having load issues.  If hashing power goes up exponentially, it will kill the pool.

Ah, good point. I didn't even consider bandwidth.

Does anybody know how much bandwidth it takes to run a good sized pool (400-500GH/s) as is?

[siggy]

I am also worried about the massive increase in power bringing pools down.  Eclipse, which I'm a member of, is already having load issues.  If hashing power goes up exponentially, it will kill the pool.

Ah, good point. I didn't even consider bandwidth.

Does anybody know how much bandwidth it takes to run a good sized pool (400-500GH/s) as is?

I've been pondering this..  Everyone is saying ASICs are going to cause even more centralization of mining... I'm thinking just the oposite..  Not only is a 10-fold increase in hash rate going to play havoc on the pools, but even on the miner side, it will not be pretty.  At approx 10 Ghash, I'm running a constant 30 Kbps.  Currently my ISP still considers that residential..  However, up that 10x to 300, and I'm not so sure..  and that just gets to 100 Ghash.. Once you get to the Thash level, you are talking constant megabit traffic.  At this point, bandwidth is no longer not-insignificant. Serious miners will to have to decide:  Stay in the mining pools for no-variance VS. Going back to mining on your own to save bandwidth.

Sigg

[nedbert9]

I am also worried about the massive increase in power bringing pools down.  Eclipse, which I'm a member of, is already having load issues.  If hashing power goes up exponentially, it will kill the pool.

Ah, good point. I didn't even consider bandwidth.

Does anybody know how much bandwidth it takes to run a good sized pool (400-500GH/s) as is?

I've been pondering this..  Everyone is saying ASICs are going to cause even more centralization of mining... I'm thinking just the oposite..  Not only is a 10-fold increase in hash rate going to play havoc on the pools, but even on the miner side, it will not be pretty.  At approx 10 Ghash, I'm running a constant 30 Kbps.  Currently my ISP still considers that residential..  However, up that 10x to 300, and I'm not so sure..  and that just gets to 100 Ghash.. Once you get to the Thash level, you are talking constant megabit traffic.  At this point, bandwidth is no longer not-insignificant. Serious miners will to have to decide:  Stay in the mining pools for no-variance VS. Going back to mining on your own to save bandwidth.

Sigg

One of the mechanisms of pools is to assign lower difficulty work, in relation to the actual difficulty e.g. 2.2m, to it's miners.  That is essentially why mining in a pool incurs less variance.

The pool operators can increase the share difficulty assigned to miners thus reducing the bandwidth requirement.  EMC pool is already testing increased difficulty shares.

[Korbman]

One of the mechanisms of pools is to assign lower difficulty work, in relation to the actual difficulty e.g. 2.2m, to it's miners.  That is essentially why mining in a pool incurs less variance.

The pool operators can increase the share difficulty assigned to miners thus reducing the bandwidth requirement.  EMC pool is already testing increased difficulty shares.

Hmm...wouldn't there be tons more stale shares as difficulty per miner increased, not to mention the new theoretical speeds we'll be mining at? Or am I thinking about this the wrong way...

[nedbert9]

One of the mechanisms of pools is to assign lower difficulty work, in relation to the actual difficulty e.g. 2.2m, to it's miners.  That is essentially why mining in a pool incurs less variance.

The pool operators can increase the share difficulty assigned to miners thus reducing the bandwidth requirement.  EMC pool is already testing increased difficulty shares.

Hmm...wouldn't there be tons more stale shares as difficulty per miner increased, not to mention the new theoretical speeds we'll be mining at? Or am I thinking about this the wrong way...

Yes, you are thinking correctly.  The significance of increased stale shares depends on the individual miner's hashrate compared to share difficulty. 

I have yet to see a comprehensive study of stales at varying difficulty.  Though, it has been said that difficulty of 5 performs well for BFL mini rigs.

[siggy]

Yes, you are thinking correctly.  The significance of increased stale shares depends on the individual miner's hashrate compared to share difficulty. 

I have yet to see a comprehensive study of stales at varying difficulty.  Though, it has been said that difficulty of 5 performs well for BFL mini rigs.

Interesting.. I did not know that there was already a mechanism in place for increased difficulty shares.  Do all the common miners already support this?   

I don't see how an increased difficulty would really have that much of an effect on stales.  Stales should only occur when a new block is started right?  As pretty much everybody already uses long-polling, I don't see how the difficulty of a share would affect how long it takes between the new block event on the server and the long-poll/new block recieve event on the miner client.  All things besides share difficulty being the same, you are going to waste just as much electricity between the two.

Sigg

[Gabi]

p2pool would solve that problem

and if you spent 30.000$ then you better be able to use p2pool instead of using deepbit.

[JoelKatz]

So long as there's no shortage of supply of ASICs, it stands to reason that they will keep selling until they raise the difficulty to the point where they no longer provide an attractive ROI. (Of course, you have to factor in risk, and there's lots of that because of difficulty changes and exchange rate changes.)

[Raize]

Sure, very profitable with YEAR roi.

After a year or less or with increasing price, every BTC mined is profit (minus the low electricity costs), so yes, it's profitable. Remember that ASIC has to actually come out. Without the power requirements from the final phase of tweaking, you can be assured BFL is still easily over a month from release testing for pool ops and coders, even. Given the way they are going about it, there's no way we'll see ASIC devices actually shipped to customers by the end of October. Also, those with GPUs have an asset that can be resold to recoup their costs, so their ROI is even lower, technically.

With regards to "YEAR"... Starting a "real world" mining operation requires loans with a repayment schedule of 3 years, and you're not really expected to turn a profit at all the first three years since you'll mostly be covering loan payments and employee salaries. But when you are done you own the equipment outright and every year after that is profitable. My point here is that you could get a 3-year loan for FPGA devices (or ASIC, once it's released) and be profitable in the first year. That's why it's silly to be concerned about a 51% attack while the price is this high.

[fireduck]

Interesting.. I did not know that there was already a mechanism in place for increased difficulty shares.  Do all the common miners already support this?   

There is a mechanism for it, a getwork request returns a target difficulty and all the software I've tested accepts it and does the right thing.  My mining pool is the only one I know of that allows miners to select their difficulty (but I don't expect this to be true for long, others will almost certainly do it).

http://hhtt.1209k.com/

[nederhoed]

Hi all and thanks to Korbman for the analysis,

Here some of my thoughts on this matter.

An FPGA running at 800 MH/s would cost around $800 running at 80 Watt.
The BFL ASIC SC presumably running 40 GH/s costs $1300 running at 100 Watt (uncertain).

While neglecting the difference in electricity consumption, the ratio ASIC to FPGA would be:
(40000/1300)/(800/800) = 30.8 times the capacity per dollar on purchase.

If we assume the current network capacity of 20TH/s is running on FPGA and invested hardware would be completely replaced per dollar by ASIC, the resulting network capacity would be around 600 TH/s.

BFL would need to sell 15000 SC's (or 600 SC rigs) for the miners to achieve this enormous figure.

Using TP's Bitcoin calculator, I would reach break even after almost 500 days:
http://bit.ly/P8lEME


Having said that. This figure will probably not be reached in a short period. Miners have to invest in hardware again. Miners have to ensure more bandwidth. Making the entire operation more like running a business.
BFL probably won't be able to ship 600 rigs, or the equivalent in single SC's in a short period. Resulting in sales for the amount of 600*$30k = $18M.


Then again, I think Korbman's TH/s prediction (143 TH/s) will be reached in 2013 easily.


Challenge me :-)
http://betsofbitco.in/item?id=612 (difficulty will exceed 8000000 on March 1st, 2013)