Topic: Bounty: development of efficient open-source FPGA/ASIC mining solution (Read 10605 times)

No bounty left unpaid?

If the benefit of the work is open to the whole community and designers(who already have the software) have the freedom to tweak it then i think its pretty much open source... Tough as it already is, including the price of the software means dropping the bounty altogether...

Riiiiight:

Sell 1000x ASIC bundles for $1k-$10k

OR

Mine for 6-months and make 10x that much money.

Nobody in their right mind would choose to sell chips at this stage of the game.  Maybe in a year or so when the reward goes to 25, or some other external factor changes the dynamics.  But there are very few Bitcoin miners able to afford paying for an ASIC array, and there's too little profit to be made selling the chips themselves.  Now and for the foreseeable future there is zero economic incentive to actually sell chips capable of mining in the GH range.  Having a tactical advantage of using ASICs isn't worth selling even some of them, since a first-mover of ASICs would want to keep that leverage for as long as possible (which dis-incentivizes mining AND selling chips). 

I could only see public sale of ASICs happening if it became known that there was more there was more than one ASIC implementation in existence, and the advantage of being first was lost.  Then it might make sense to start selling publicly.

Those factors notwithstanding, I applaud the efforts of the community to find a viable implementation.  However, there are 2 big problems facing the community that have, as far as I'm aware, no solution:
1. ASIC design engineers are very scarce (especially ones who really know how to do more than one part of the design flow)
2. Providing funding for this project is very difficult due to the up front NRE.

The only people I've ever heard willing and able enough to overcome this are:
1. Artforz (who has a private implementation)
2. The Canadian guys (They claim to have a team working on this, but I can't find their posts.  Search and you will find them.)

Anyways, good luck to everyone on this project.  It would be nice to see a  publicly accessible solution.

EVERYTHING has a price.

There is an issue that worries some of us at the modular miner thread, and I think it is highly relevant here. Many of the better FPGAs require paid-for software licenses in order to develop any code. For example, the modular miner thread has agreed on a chip that needs $2000 a year for such a license.

It is debatable whether this makes the project any less "open source"; after all, even the $0 software for FPGA development is closed source. However, my question is, do we include such a license fee in the Mhash/$ criterion?

It may be possible that other people can use the generated design without the software, so we should only factor in the hardware cost. However, in that case I would not consider the whole package "open source", because I would not be able to play with the code at that price.

I generally refuse to buy any hardware that I cannot program freely, and I think this is particularly important when buying something for a singular purpose. Even if Bitcoin fails in the near future, I can continue to use my GPUs and FPGAs in other fun ways.

I don't see either. I'm just pointing out that it's something they could do. I can't defend the claim that there are no imaginable vulnerabilities.

The chip could still silently discard all found hashes that exceed a particular difficulty level. Testing with difficulty 1 work, you'd be none the wiser. Eventually, people using the ASIC would be getting shares from pools but not finding blocks.

Now that I've pointed out this specific threat, it's obvious how to test an ASIC for it.

I don't see what benefit that would have to the manufacturer, unless they created a competing company that sucked up all the business after the "scam" was discovered, essentially getting people to buy the same chip twice ... which is actually ingenious. MWAHAHAHA

Anyway, you'd just feed the chip Difficulty 1 work to work-around the "feature". The chip should be operating on Difficulty 1 work anyway, because there's no reason to have the extra logic in the chip to check for other difficulties.

There are mechanisms available that would allow a person to have confidence that the ASIC did what it claimed. In operation, the inputs and outputs of the ASIC would be completely controlled by software under the user's control.

Essentially, the ASIC is like a miner and the user is like a pool. Controlling the ability of a miner to harm a pool, ensuring the miner is doing what they claim to do, and the like are all well-understood problems with known solutions. It is trivial to confirm that the ASIC is doing what it claims to do, and there is no secret communication channel between the ASIC and its manufacturer.

The most the manufacturer could do is something like designing the chips to fail to ever find solutions that exceed a particular difficulty. To check for these kinds of defects, someone would have to imagine each particular subtle flaw and test for it.

My questions have been addressed on the (now unlocked) ASIC dev thread.

I was referring to block chain integrity, not that of my personal computer. The current clients are open sourced to enable oversight. One dominant chip manufacturer could effectively have a back door to a large portion of the network.

This isn't such a paranoid concept if you consider where many semiconductors and boards would be manufactured. China already has laws against using digital currency for material goods. There are a lot of places in the manufacturing process where dishonest changes could be made. For this reason, some agencies have their own fabrication facilities.

The previous poster wasn't really addressing my concern. I most likely wouldn't be controlling ASIC arrays from a PC.

I don't claim to be a VLSI expert, I'm merely inquiring.

Dude, if you're worried about the evil ASICs taking over your computer, stick them in a external chassis and refuse to talk to them except through a USB port. With a large enough block size the communication between the ASICs and the host computer is minimal, even if you put a hundred of them in the box.

If the host computer has a compromised Ethernet interface with Direct Memory Access, it doesn't matter.

I realize it is not practical to wait 150 years to experiment with a crypto-currency, but I still think Bitcoin will fail medium-term. Hopefully valuable lessons will be learned for bitcoin 2.0.

That is of course moon-man talk, unless you're planning on giving the ASIC its own Ethernet interface.

The problem is deeper than you realize. Modern computers are inherently insecure, and since at least 1996 software has been designed to work against the user. In the past decade, more and more hardware has been designed to work against the user as well.

Even if you trust Intel and AMD not to add a back-door that phones home for instructions upon seeing a specific 256bit number, do you trust the Chinese (or American) government not to tamper with the design and manufacture of those components?

Even if such tampering can be found, it can likely be explained away as an oversight or design defect; especially if the mechanism is subtle. Currently, no computer manufacturer will absolutely guarantee their systems will work as advertised. With GPUs, the full specs are not even publicly available. Building trusted, proven correct computers will likely take generations (like 150 years). If a computer is proven to implement a certain specification, you can sue if any "bugs" are found: since the design is 'proven correct', any 'bug' must be deliberate sabotage.

Seems the latest ASIC thread has been locked, for some reason.
 http://forum.bitcoin.org/index.php?topic=14910.0

How would you audit a project like that for security? If enough miners adopted the same ASIC, a >50% hashing power problem could present itself. It seems you'd have to trust the vendor.

Me too. Power efficiency is much more important in the long run.

I'll move my 20 BTC bounty to the $2.00 price point. I think an open source approximation of ArtForz's capability is a good starting point. 

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

Unfortunately I'm not in a financial position to be offering bounties at the moment.  I withdraw the bounties that I offered on 5/17/11 and 5/24/11 until further notice.

- -Chris Acheson, 5/27/11
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
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=Gdgq
-----END PGP SIGNATURE-----

I picked that requirement as a "competitive with GPU mining" price point.  I think I might be interested in investing in FPGA mining around a $2 per MH price point, so I'll add a second bounty for that.  Unless I'm misunderstanding what he said, ArtForz has already achieved this:


It would be good the have bounties at various development milestones anyway.  If there's anyone else who wants to change their bounty requirement, let me know and I'll update the original post.


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

I will offer an additional 10 BTC bounty with the same requirements as my previous bounty offered on 5/17/11, with the exception that the device must cost no more than $2.00 USD per megahash per second.

- -Chris Acheson, 5/24/11
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
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=OCTV
-----END PGP SIGNATURE-----

It will be pretty hard to get below $0.60 USD with FPGAs, I really had the ASIC alternative in mind for my part of the bounty, but I will of course pay it for a FPGA solution that meet all requirements.

A full implementation exists:

http://forum.bitcoin.org/index.php?topic=9047.0
https://github.com/progranism/Open-Source-FPGA-Bitcoin-Miner

It's fully functional, ready to go. There's even a binary release, which is pretty easy to use. I wrote scripts to program and control the board, so you don't even need to know how an FPGA works to use this.

I guess it doesn't meet the bounty, though Because of this

It easily meets the Watts per MH/s requirement. I measured 8Watts at the wall for the 50 MH/s design  and that's with a dev kit! But it obviously doesn't even come close to the $0.60 USD per MH/s requirement.

But I will continue to work and improve on it. The $0.60 USD requirement is insane, but so am I! 

LOL

I dumped some vhdl-code on this thread half a year ago

http://forum.bitcoin.org/index.php?topic=2362.0

It runs with ghdl.

It is synthable but you'd need to add a counter for the nonce and a comparer for the hash value to be useful.

Also the host communication is not imlemented.

Pipelining would also do good for performance.



The code is translated into vhdl from the free verilog implementation at opencores.org

Maybe this will get someone closer to a full implementation.

You can't make that claim.  You don't know this.  Also, you don't know what others pay even for power.
IF they bought it to play games, yes they do.
They probably wouldn't have, but some will now.
Just because you can't imagine it, doesn't make it wrong.  Again, you need to ask for your econ education funds back.

There is a point for any of these guys that continued expansion of their own cluster becomes a case of diminishing returns.  So at some point, they will stop expanding.  I didn't say they would sell the cards they already have installed, but the IP of their design.

Supposedly ArtForz has already developed a custom run of asic based pci cards for his mining cluster, he just hasn't released his design.

This is a ridiculous statement.  The costs for a pro miner, per MH, is far lower than an individual miner, purely due to efficiencies of large-scale mining.  Everyone pays for the GPUs, everyone pays for power.

Just because someone bought the card for gaming doesn't mean they get to automatically ignore it as a cost for mining.  Plus these are not people who will buy the ASICs anyways.

I can't see this happening.  At the time when inflation drops off to low levels, transactions processing fees will be more than adequate to make up the difference.

Anyone with enough capital to fabricate an ASIC will require maximum return on investment.  Maximum return will always be mining, and in later years fee processing.  It will never ever be eclipsed by selling chips as commodities.

I have been seeing a lot of times comments that make it sound as if a US$300,000 development cost for an ASIC is a crazy-high cost, far out of reach.

Yet I also recall having read this same day that the bounty for an open source GPU miner was BTC50,000.

How much is BTC50000 nowadays?

Isn't it actually *more* than US$300000 ?

So whoever picked up that previous BTC50000 bounty might single-handeldy be in a postion to be able to afford to have an ASIC developed?

How many others have BTC50000+ lying around, maybe from back when they picked it up for a penny a coin or so?

Surely a mere US$300000 to US$500000 or so is far from out of reach for the bitcoin community?

-MarkM-

This is an issue, to be sure.  However, ASIC's can be mass produced on a level vastly more cost effective than equvilant FPGA's.  So much so, that if the crypto is changed, it would still be more cost effective to buy a second ASIC mining card than to buy one FPGA card.

ASICs are also retardedly fast, compared to FPGAs.

You need to return to your Econ prof and ask for your money back.  You are overlooking some important variables.  First off, professional mining cannot compete over the long term with individual miners, who have near zero facilities costs.  Some don't even have capital costs, because they are presently using GPU's that they would have bought regardless.  Eventually, Purpose made bitcoin mining cards will become available.  For no other reason than once all those heavily vested professional miners (such as ArtForz) reach a point that their profit margins are squeezed too much to continue to expand, they will monetize their hardware designs by selling the cards for individual users.

I would think that any mining "board" should be built with a FPGA. ASICs can't be changed at a later date, so if bitcoins ever switch from SHA-256 to a different crypto all the boards would be useless. With a FPGA, you have a better chance of being able to program the new implementation.

The way I see it for mining, difficulty and income rates will eventually reach an equilibrium, where only people with extremely efficient MH per second per Watt rates will make a profit. And if open FPGA/ASIC developing takes off I'd imagine that's where at least part of the large scale mining will head, due to the power efficiency of this solution.  That is, of course, if nothing happens to crash the BTC value to sub-dollar levels and/or to make people use BTC less or not at all.

And speaking of hardware, when FPGA will be a more manageable solution It would be great if someone will start trading/shipping such hardware and offer configuration services for BTC...

This would be a great idea if somebody can develop specialized bitcoin mining hardware.

We are making too much money for greedy ATI at the moment. Nvidia maybe you are interested ?

10 BTC here.

The numbers in the bounty requirements that I posted are a marginal increase over the efficiency of existing GPUs.  The power consumption requirement of 0.3 watts per megahash (as compared to about 0.4 w/mh for a 5970) is pretty much a freebie.  The dollars per megahash requirement of $0.60 is not even as good as what you get from a 5850 (mine were 50 cents/mh at $170 each, including express shipping), but is in the same ballpark as a decent price on a 5870.

The point of this bounty is not to entice someone to design and give away the ultimate ASIC that's going to render GPU mining totally obsolete.  The point is to encourage someone (perhaps a student rather than a full-time professional engineer) to work on optimizing an FPGA design to the point where it's merely on-par with current GPU mining; and to share that design with the community as a starting point for continued open development.

This bounty is much more likely to be achieved with FPGAs, but I figure there's no point in excluding an unforseen ASIC-based solution from it.

Artforz has stated that his ASICs, while quite power-efficient, are not competitive with GPUs in terms of dollars per megahash.

I hate to be the one to be the destroyer of hopes & dreams, but I need to explain the economics of this to you guys.

Nobody will ever sell you an ASIC miner.  The reason is simple: profits.  The amount of money to be made selling ASIC is so miniscule compared to the profits that will be made by using them for mining.  Using simple estimates, I would say it is 1000x more profitable to use ASICs for personal mining vs. selling them.  It makes no sense to sell them to other miners, which only increases the global difficulty anyways.  Furthermore, the number of savvy miners around who would buy ASICs is small (I'm guessing less than 100 at the time of this post), so the market is tiny.

The only point at which it would make any financial sense to sell them is after the block reward goes down to something really small.  At that point, mining would have very limited profitability, and money would be made from transaction fees.  Then it may make sense to sell them to other miners.

Just as a reminder to everyone, Artforz already has mining ASICs.  He is wisely keeping them to himself.  This milestone already has been reached, and the number of people who will join this club is very very limited.

I too will pledge 20 BTC for the satisfaction of the conditions laid out by cacheson in the original post.

20 more BTC here.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Great initiative! I've actually been researching ASICs a few times, but somewhere on the way I realized that I'm more interested in Bitcoin success than increasing my own holdings, so open-source ASICs sounds great.

I'm contributing 20 BTC.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (MingW32)

iQEcBAEBAgAGBQJN0vFCAAoJEJ8xgCx5ZC8lH2AIAK9YJfncAqZ6Ek/WUClLJLb9
KyiQjNB3RgLWumX390Jf4Wn5G6CHNEIXiLXOg0DSCnzoY2y0LtLznJ6kigU8mEaU
kw7XHH0+u5wF4d+tigz1RXNY8vAoIHQFhhT/aeNdoZCk6x2Gata+iX5eQ79xzvmU
0Ili/EAJHoMVDQODf+fqXGiVU1Zw2Fb8xwZ0d3jNLsQet1U9kEiF+cAAf/FxaRns
Rhk6M4OpOJzorvnXkcqquGDDD0S1UbfuQ07mWZON1JREh8SR1j3DsVdZ12rsO+K/
ycbD2Yo1kNV+9eWHbCm5JvbOquEXGyloL7tmcbWdBuHz6HxOSQEhX4gWaGkxbNY=
=vcWF
-----END PGP SIGNATURE-----

I'm splitting this off from the FPGA mining for fun and profit thread.

Current contributors:

$2.00 USD per MH/s and 0.3w per MH/s:

• BitterTea - 20 BTC
• teknohog - 10 BTC
• cacheson - 10 BTC retracted

Milestone Total - 30 BTC

$0.60 USD per MH/s and 0.3w per MH/s:

• cacheson - 10 BTC retracted
• gmaxwell - 10 BTC
• riX - 20 BTC
• gusti - 20 BTC

Milestone Total - 50 BTC

Grand Total - 80 BTC

Those of you who are interested in making sure that Bitcoin remains a distributed system through the transition from GPU mining to FPGA/ASIC mining are encouraged to contribute.  I know the total isn't much yet, but if enough of us chip in we should be able to provide a decent incentive.