DESIGNING the next generation FAST CRYPTO CURRENCY MINING MACHINE

[Viceroy]

Last edited:  June 9, 2013 (refresh to check for update)


Ready to DESIGN, NOW seeking advice on development boards for a litecoin mining FPGA.




Now Recruiting a DESIGN Team

It appears the ASIC race just ended and Avalon won. Avalon shipped completed 65GH/s mining units and has taken orders for at least an additional 50,000 of their 282 MH/s bitcoin mining chips.  Butterfly labs has shipped their very first 5GH/s unit with more anticipated to follow.  The difficulty curve is about to jump to 140,000,000 and soon thereafter rise to 700,000,000.  At that rate my personal (theoretical) 3.6 GHz farm produces pennies a day and the cost of electricity is far more than I can generate in coin (assuming a $100 valuation for Bitcoin).

-- to keep the thread readable, please minimize your use of the quote function --

So, what's next?

I have come to the conclusion, as many others have, that the future of Bitcoin died with the shipping of the ASIC.  Satoshi's original paper starts with a description of a peer-to-peer currency yet the ASIC mining solution destroys the peer-to-peer methodology and places all the power of printing Bitcoin into the hands of a tiny group of companies who have the capital to produce a massive farm.  To compete against these farms you need even more capital.  

Per the calculator, if one bitcoin is worth $100US then:  

@140,000,000 a 1GH/s miner yeilds $0.34US per day before electricity
@700,000,000 a 1GH/s miner yeilds $0.07US per day before electricity

So what can be done?  What if I can build a farm that operates at 1 tera hash per second?

One Avalon operates at 282MH/s so I need about 3500 chips at $20k+ (plus the multi-thousands dollar PCB design/production).  Or I could buy 15 prebuilt units at for approximately $200k.  So if I budget $250,000 to a tera hashing solution and plug it into the calculator I find:

@140,000,000 a 1TH/s miner yeilds $333US per day before electricity (~$500)
@700,000,000 a 1TH/s miner yeilds $72 US per day before electricity (~$500)

Gee, this isn't going well is it?

So what can be done?  What if I can build a farm that operates at 1 peta hash per second?

We need a whole new animal to get to this level.  We'll need about a billion transistors and our cost to develop will skyrocket to $50 million minimum... Intel spends BILLIONS developing chips like these.  I don't think your average miner is going to get there.

And this is why Bitcoin will die.  

What we need is a new crypto currency based upon the Bitcoin design which does two things:  
1) the coin needs to work based on an algorithm other than sha256 since ASICs do the sha256 proof of work amazingly well.  
2) the currency should show transactions more quickly than bitcoin.  (Can you imagine paying for something at a store and needing to wait 30 minutes for the cash to end up in the register?)  

I propose, because miners can no longer mine with their personal computer and miners are the heart of the currency that wise miners will migrate away from bitcoin to litecoin because it solves these problems.  Further I think cryptocurrencies are still in their infancy with mass adoption still several years, if not decades, off.  So where is the opportunity?

I believe in bitcoin for a variety of reasons and I am long term bullish on crypto currencies.  But I also believe people will continue to migrate to scrypt based currencies because their fancy GPU mining hardware is now useless in solving the sha256 proof of work.  So I would recommend to anyone they point their gear at litecoin (though that doesn't help me).   Then we get into a theoretical question about scrypt.  It is designed to be ASIC hostile to twart groups like Avalon from centralizing mining solutions.

So I'd like to begin a discussion about building a device that solves the scrypt proof of work.  And so we need to start with a rudminetary understanding. Here it is in all it's wikipedia glory

"The scrypt function is specifically designed to hinder [brute force attacks] by raising the resource demands of the algorithm. Specifically, the algorithm is designed to use a large amount of memory compared to other password-based KDFs, making the size and the cost of a hardware implementation much more expensive, and therefore limiting the amount of paralleling an attacker can use (for a given amount of financial resources)".

So where is the opportunity?
Who wants to help me try to CENTRALIZE the scrypt mining process?

-- to keep the thread readable, please minimize your use of the quote function --

note:
At this time this is a volunteer effort, nobody is getting paid for anything.  
We are seeking engineers and mathematics, coders and builders.  
We are not seeking investors, though we invite them to watch the process.



The original Scrypt whitepaper:

Interesting thread. I had a brief look at the original Scrypt paper (which is quite old)

    


    


older notes that led to this conclusion:
Recovering Investment:

Analysis of Difficulty curve

Seems like getting back your sunk cost will be risky. BFLs vaporware is at $50/Ghash, avalon less if the DIY option materializes. $75+profit+board costs seems high in comparison for only 600 Mhash (assuming your estimates are accurate).

Other ASIC makers are getting to the point where they can produce large quantities and at that price you might not get much volume but I'd expect you'd get a fair amount of support from your local market if the mh/w is comparable.

Xilinx Spartan-6 price delivering ~230 MH/s at 25k Quantity is $50.....@5k quantity is $72.....6 weeks lead time...for 10 weeks lead time you get ~280 MH/s @~$9 - AVALON.... Everybody in this forum talks about ASICS and NRE of ~250k (where the ETA is ~ 4-5 months) and nobody buys 10k Avalon chips....it's a lot more easier and cheaper

You may be right, but until I have a delivery schedule and actual price in front of me I am betting there may be a market.
Costs & Cost Reduction:
I've sent off the Open Source FPGA Miner HDL to analyse it and determine a cost to produce it as an ASIC.  I recognize that changes need to be made to convert it but I don't posess an ASIC RTL at the moment so this is the starting point.

You are looking at a cost close to 2M usd, at least. Keep in mind that the upfront cost of design/production increases with a super linear scale with finer process node. 110nm process node (avalon) was easier than 90nm. 28nm will be much much harder.

You mentioned the price depends on volume, how much volume needed for the USD 65 to 85 mark and how much to get them at half that price?

Collaboration:

Put up a GitHub to evaluate the design you sent out to check.

From the Open-Source-FPGA-Bitcoin-Miner project:
https://github.com/progranism/Open-Source-FPGA-Bitcoin-Miner/blob/master/projects/DE2_115_makomk_serial/fpgaminer_top.v
https://github.com/progranism/Open-Source-FPGA-Bitcoin-Miner/blob/master/projects/DE2_115_makomk_serial/sha256_transform.v
https://github.com/progranism/Open-Source-FPGA-Bitcoin-Miner/blob/master/projects/VHDL_StratixIV_OrphanedGland/sha256/rtl/sha256_pc.vhd
https://github.com/progranism/Open-Source-FPGA-Bitcoin-Miner/blob/master/projects/VHDL_StratixIV_OrphanedGland/sha256/rtl/sha256_qp.vhd
Technology:

Viceroy, what kind of throughput are you trying to develop?

Are you building ASIC or FPGA? ASIC does not have LUT (look-up table), it has gates.

Happy to consider ANY technology that will deliver what BFL has not.  If you have directions to build a quantum computer please send them my way  
The Manufacturing Process:

Typical ASIC production will involve:
1. HDL design from spec (100 - 200K)
2. Verification & physical synthesis/timing closure/power analysis of (1) (>200K)
3. Contract with fab and production (> 1M)
You can slice and dice above based on which components you get ready to use. Since you would be targeting a lower process node, likely other than (1), you will need to pay for other points, and expect the iteration between 2 and 1 very high (implying need to pay for 1 as well).

About Viceroy:

Now out of curiosity, do you have experience in the field? Or is this just your first design?

I am a computer scientist educated at a University.  I am not an electrical engineer and do not have the ability to develop this chip.  I do, however, have decades of management experience with and around engineers.  In terms of project management I have designed and managed an online database of Dept. of Education data through four revisions while operating as a strategic revenue sharing partner with several major universities and several non-university online publications.

[Viceroy]

Come on Inaba, it's ON!  

Well... that was fast...  Glad I didn't "bet".  

Here's the latest news: Coding In My Sleep Receives BFL on 420 2013

This is a significant event and, ironically, I think it spells the end of Bitcoin.  What it will do is force the 'average user' to migrate to a currency like litecoin where the miners work based on scrypt instead of sha-256.  Now the problem, and the opportunity, has realistically just shifted to scrypt mining.  Your 'average joe' will no longer mine bitcoin and it will slowly fall out of favor.

Disagree?  Opportunity?

[Viceroy]

Agreed 200%.

Thank you for that.  I agree 200% as well.  Another nice thing about litecoin is that the transaction takes 2 minutes to verify as I understand it... not 30+ so real merchants are more likely to adopt it (imagine waiting 1/2 hour to pay for gasoline!).  Finally they offer more coins with 84 million total compared to the 21.  (I think I would have chosen a much much higher number). 

So if you guys are all in let's start figuring out what a scrypt miner looks like....

[Viceroy]

Interested. Watching thread.

Would this be a ASICMINER-esque operation or a Hardware Sales operation (such as Avalon)?

Whoa slow down there, camper, let's see what we're gonna design before we come up with a way to sell it.  ;-)

[Inaba]

While I actually support a project like this and I hope it (or something very much like it) succeeds, I find it somewhat funny that you claim to be "ASIC HOSTILE - Decentralized mining is the only workable solution" in your signature while also proposing building a mining ASIC along the same lines of Avalon and BFL.  Cognitive Dissonance much?

[JoelKatz]

I disagree with your for two related reasons:

1) Crypto-currencies that are most efficiently mined with ASICs are actually more secure because general purpose computing equipment can't be re-tasked to attack the currency. Once Bitcoin is mined with ASICs, retasking a botnet to attack Bitcoin will be hopeless.

2) Crypto-currencies that are most efficiently mined with ASICs can only be attacked by people who have a significant investment in equipment whose value is directly determined by the value of the currency. Those who can mine Bitcoins efficiently will be those who have invested heavily in Bitcoin itself and will have the most incentive to preserve its value.

There are a number of possible attacks on Bitcoin based on having lots of mining power, but the fact that it can be heavily accelerated by ASICs actually *increases* the resistance to them.

[Viceroy]

How then does the 'average Joe' at home get involved without needing to buy specialized hardware?  doesn't forcing people to buy asics destroy the decentralization in favor of large farms?  do you not expect these now burned early adopters to migrate to another algorithm and start spreading the word of a better/faster crypto currency?

And is my math right?  we need a billion transistors and $50 million to build a terra hashing sha chip, right?

[Vagnavs]

sure a litecoin project interests me. With the latest developments, switching over sounds a good idea. You make some very good points as well.
Regards,
Brian

[Viceroy]

The way i see the Op:

1.Bitcoin will die because of the ASIC invasion.
2.People will move to an asic-hostile cypto currency most likely litecoin.
3.Help me be among the first to build a litecoin ASIC mining machine so we can kill Litecoin as well while making some money in the way.

Bonus:Op not only predicts a shift towards litecoin mining but also recommends people "to point their gear to litecoins" because they are ASIC-hostile.

 

yes but 3 only works if 1 and 2 are true.  is there something wrong with this line of question in an attempt to learn?

[JoelKatz]

How then does the 'average Joe' at home get involved without needing to buy specialized hardware?

They don't. But nothing stops the average Joe from buying specialized hardware.

doesn't forcing people to buy asics destroy the decentralization in favor of large farms?

No. Small farms can use ASICs too.

do you not expect these now burned early adopters to migrate to another algorithm and start spreading the word of a better/faster crypto currency?

If there's a better/faster crypto-currency, then Bitcoin deserves to lose to it. However, ASIC-based mining is a strength, not a weakness.

And is my math right?  we need a billion transistors and $50 million to build a terra hashing sha chip, right?

I haven't done the math, but that seems reasonable.

[malevolent]

It should take 10 minutes with a transaction fee that isn't too low and with a few measures* instant 0-confirmation transaction can be relatively safe and used for transactions of, say, less than $1000 (given the resources required to double-spend if appropriate precautions were taken by the merchant, accepting 0-confirmation transaction for low-value goods can be safe).

*connections only to previously whitelisted and known honest nodes, no outgoing connections (UPnP off), etc.

[Viceroy]

It should take 10 minutes with a transaction fee

Not sure where you are from but 10 minutes waiting at a cash register is unacceptable in American commerce.  So if you are third in line it will take 1/2 hour to buy milk.

[MooC Tals]

It should take 10 minutes with a transaction fee that isn't too low and with a few measures* instant 0-confirmation transaction can be relatively safe and used for transactions of, say, less than $1000 (given the resources required to double-spend if appropriate precautions were taken by the merchant, accepting 0-confirmation transaction for low-value goods can be safe).

*connections only to previously whitelisted and known honest nodes, no outgoing connections (UPnP off), etc.

Lets not forget the other strength of scrypt base crypto called litecoin. The difficulty re-target is much more quicker. I always have seen litecoin as the little brother of bitcoin but if this hypothesis is true then litecoin could be come the new bitcoin. I can seethis happening if the barrier of asics not being able to mine litecoin is circumvented. Litecoin was considered gpu proof at one time.

[crazy_rabbit]

I don't quite get the point about why ASIC's kill bitcoin. Mining was always going to trend towards being a razor thin profit business where only the largest make any sort of real profit (once the price of bitcoin stabilizes-  if it goes to $10.000 each then these calculations are a little off  )

Its true there is a big opening at the moment for ASIC companies to centralize mining- and they will indeed. But companies like BFL seem like they are focused on getting as many ASICS into the hands of the general public as possible. We all knew that ASIC's were only printing gold if you were the first one to get one. GPU farms prior to ASICS managed in some fashion to centralize mining. Someone with a thousand cards made far more then someone with ten. But it didn't stop anyone. Now we just all have to buy asics and we have to give up on the argument of "if bitcoin fails at least you have a sweet GPU for gaming).

I don't see how things have changed. The landscape has restarted and we all need to buy ASICs. I surely will, even if it's not the money-printer I was hoping that it would be.

Building ASICS for Scrypt is interesting, and it's only a matter of time (if litecoins price rises at least). I'll stay tuned.

[malevolent]

It should take 10 minutes with a transaction fee

Not sure where you are from but 10 minutes waiting at a cash register is unacceptable in American commerce.  So if you are third in line it will take 1/2 hour to buy milk.

Not sure where you are from but reading doesn't hurt For low-value transactions no waiting is necessary.

So I'd like to begin a discussion about building a device that solves the scrypt proof of work.  And so we need to start with a rudminetary understanding. Here it is in all it's wikipedia glory

Are you sure it's worth it? I like scrypt but it is still relatively new and I wouldn't be surprised if someone found a flaw in it that would render it less safe, after all it is as old as Bitcoin (created in 2009) and also created by only 1 guy (although Satoshi did use some other peoples' ideas and consulted them). If the flaw requires a change of an algorithm in Litecoin, all those expensive ASICs will become useless.

It will be a long before it will be worth to invest so much money to design & manufacture an ASIC for scrypt.

We all knew that ASIC's were only printing gold if you were the first one to get one. GPU farms prior to ASICS managed in some fashion to centralize mining. Someone with a thousand cards made far more then someone with ten. But it didn't stop anyone. I don't see how things have changed. The landscape has restarted and we all need to buy ASICs. I surely will, even if it's not the money-printer I was hoping that it would be.

+1
As Bitcoin gained in popularity the barrier for 'an average Joe' became too high already without any ASICs. Until recent price increases a lot of people wouldn't have been able to afford to mine with their electricity costs and often hardware costs (in US GPUs are cheap, especially 2nd hand). Difficulty is growing and when it stabilizes with price, only those with the best economies of scale will remain in business with profit margins decreasing.

[Viceroy]

Not sure where you are from but reading doesn't hurt For low-value transactions no waiting is necessary.

Experience.  "Bitcoin transactions are instant - the confirmations aren't. It usually takes less than an hour for the first confirmation."  There is no such thing as a "low value" transaction... Milk is NOT free and if buyers use bitcoin you are asking the merchant to risk being robbed... else hold his customer for an hour.

Are you sure it's worth it?

It's CERTAINLY worth having a discussion, yes.  Particularly about this:   

I like scrypt but it is still relatively new and I wouldn't be surprised if someone found a flaw   

only those with the best economies of scale will remain in business with profit margins decreasing.

As a "rich American" I find this troubling.  People who cannot (or will not) afford ASICs will STILL want to mine.  Crypto Currency is about MASS adoption.  If only Vlad is mining then everyone else will migrate elsewhere. So Vlad mines Bitcoin and the rest of the world moves to some other currency they *think* cannot be beaten with an ASIC.

[Viceroy]

No. Small farms can use ASICs too.

I did the math above and can't figure out what a "small asic farm" is.  How many GH/s?  Or do we need to "bank" on btc shooting to the moon?

How then does the 'average Joe' at home get involved without needing to buy specialized hardware?

They don't. But nothing stops the average Joe from buying specialized hardware.

I think this is wrong.  This is a flaw in the fundamental design which forces centralization to "industrial bank-like entities" where decentralization and every person in the world mining on their own system should be the driver.  Let me ruminate on this a bit.

* Viceroy turns to the peanut gallery and whispers

I'm TRYING to figure out a "play" here.  I believe that miners, like me, will conclude that LiteCoin is a better bet for them and further I believe that people will be driven away from BFL.  Avalon is amazing and I can't wait to see what people do with their 450,000 chips that just went to press but I think the small guy will NOT buy into the ASIC concept if Inaba remains the voice of that solution.  Building an ASIC solution is simply out of the question at this point, just go buy the Avalon chips if you think that approach makes sense.  That assumes you are less well capitalized than Intel.  (Watch Joel's answers... he's smart).

[malevolent]

Experience.  "Bitcoin transactions are instant - the confirmations aren't. It usually takes less than an hour for the first confirmation."  There is no such thing as a "low value" transaction... Milk is NOT free and if buyers use bitcoin you are asking the merchant to risk being robbed... else hold his customer for an hour.

It will be long before we can expect people to use Bitcoin for groceries, but setting this aside, given precautions I mentioned few posts above the risk becomes largely mitigated and acceptable for low-value transactions (I can be struck by a lightning any time I am outside, does that mean I shouldn't leave my room?)

Are you sure it's worth it?

It's CERTAINLY worth having a discussion, yes.

I wasn't clear enough, discussion, sure, but do you really believe it's worth to design & build an ASIC for scrypt in the foreseeable future?

We are seeking engineers and mathematics, coders and builders.

only those with the best economies of scale will remain in business with profit margins decreasing.

As a "rich American" I find this logic flawed.  People who cannot (or will not) afford ASICs will STILL want to mine.  Crypto Currency is about MASS adoption.  If only Vlad is mining then everyone else will migrate elsewhere.

Gold is used widely in the world and people still use it despite the fact that only a few can afford to mine it themselves (nothing stops people from buying it). Bitcoin will someday (hopefully) become widely used like paypal but already now people who use Bitcoin actually buy it (or offer goods/services for BTC) because not everyone wants to bother/can afford/has time/is technically knowledgeable enough with setting up a mining operation (like with gold). Bitcoin, like gold, is used due to its advantages over fiat, not because ''it's a money-making machine''. The ''money'' part is only there to secure the network, Bitcoin is about financial independence from the state and middlemen.

[E3V3A]

Interesting thread. I had a brief look at the original Scrypt paper (which is quite old), but it was not obvious to what is meant with making memory intensive tasks more expensive. I'd like to see this quantified using numbers with today's memory technology. In addition it was even stated in the paper that one of the limits were based on 2-dimensional memory ingots, which we know will change in the next couple of years, to become 3-dimensional blocks of some electro-optical substrate. This is will have to effect of increasing memory size with x³ instead of x² for the same price. Now, if there's another way of mining that we could think of, that could avoid the mining growing pains into adulthood, that would be very interesting.

[uMMcQxCWELNzkt]

I could design a cool looking rig if needed, depending on components I could design to help airflow and other features needed. Below is sample of my BitPhantom mining rig design concept, I am currently working on a more lightweight consumer version. http://www.owenprescott.com/blog-bitcoin-mining-hardware.php