FPGA mining for fun and profit

[allinvain]

How much does a Cyclone4-150 board cost?

Also do you get a discount of you buy the Spartan boards in bulk?

Kinda shame you can't get more hashes out of the Spartan boards cause compared to say a cheap 5850 it gets killed. If these boards cost like $50 or something that may be worth it for 80 Mhash/s I'd definitely buy 4 to begin with - maybe more.

[1714860732]

1714860732

[1714860732]

1714860732

[caston]

How much does a Cyclone4-150 board cost?

Also do you get a discount of you buy the Spartan boards in bulk?

Kinda shame you can't get more hashes out of the Spartan boards cause compared to say a cheap 5850 it gets killed. If these boards cost like $50 or something that may be worth it for 80 Mhash/s I'd definitely buy 4 to begin with - maybe more.

I think you'll get best results if you can share the workload between the GPU and the FPGA. I don't have the technical know how to get this going but it might be possible to do it with something like the ODG1 in your mining rigs PCI slot.

http://www.linuxfund.org/projects/ogd1/

[greenlander]

I might be interested in helping out with this project with either time or money.  I'm a system software engineer with some experience writing VHDL with a EE degree.

I did my own back-of-the-envelope calculation similar to others on this forum.  I figured you could probably fit a single unrolled pipeline in a LX75 Spartan 6.  I agree with cypherf0x that it's not completely clear whether it's possible or not.  For the sake of argument, let's say it is.  Then if you could run at 100 MHz then you could get 100 MH/s.

The problem is that it doesn't seem much cheaper than GPU computing.  The Spartan 6 LX75s are about $100 each at Avnet.com (one of Xilinx's distributors).  By the time you figured in the fabrication and assembly costs you might break even in terms of "hashes per capital expense"

However, it would certainly be more power-efficient... and a lot more dense.  If you were clever you could put arrays of them on DIMM-like modules like this design: http://www.sciengines.com/copacobana/ .  You could potentially fit 100 or more FPGAs in the case that a single desktop computer fits in.

[ryepdx]

You could potentially fit 100 or more FPGAs in the case that a single desktop computer fits in.

100 FPGAs @ 100Mhz = 10Gh/s in one computer case? I would gladly pay for that. My biggest constraint right now is physical space, followed distantly by the amount of electricity my circuit breaker can route my way.

Do want. O_O

[kefir]

100 FPGAs @ 100Mhz = 10Gh/s in one computer case? I would gladly pay for that. My biggest constraint right now is physical space, followed distantly by the amount of electricity my circuit breaker can route my way.

Given that these numbers are right:

Do you think it would be worth it? If you're willing to do this, then probably a few others are too. If you buy a rig like this for US$50000 (dunno the price, just assuming something), you'd need to make ~7142 bitcoins at a value of US$7 each before it's worth it, or 143 blocks. Remember that the network only hands out 210000 chunks until 2013, so difficulty will increase as you are doing this. The question is - will it pay off? Will you actually be able to use this rig to get those 143 blocks?

Maybe smaller investments is the way to go for bitcoins (unless you already have some fancy hardware). Buy something, generate a couple of blocks on it, then buy some new hardware that is better, and generate a few blocks on that, etc... Heavy investments will probably be overtaken by bitcoin limitations and technology before they've paid off. If so, what would be the right level of investments? If someone works on it, I guess it should be possible to find an optimal price level vs. block performance ratio that could be a buyers guide to get into bitcoins.

Unless you want to gamle that bitcoin prices will increase a lot, of course. You're already gambling that they'll stay at current levels.

[caston]

I think if you want to do this make it a public mining rig. e.g. people can buy shares in it to allocate the capital for the purchase of the mining and FPGA hardware then people will get a return on these shares. The alternative is is a few greedy guys getting all the bitcoins for themselves and then everyone else  loses interest in them and moves onto the next craze. I would also like to suggest that people can be paid in shares rather than bitcoins directly for their help, elbow grease, attention and expertise they put towards helping build the company. Please also make everything such as the business plan and technical documentation and diagrams open source.

[ryepdx]

I think if you want to do this make it a public mining rig. e.g. people can buy shares in it to allocate the capital for the purchase of the mining and FPGA hardware...

+1 to everything you just said. I'm not an engineer, but I can supply some capital and some programming help if some enterprising and learned person wants to point me in the right direction.

[caston]

+1 to everything you just said. I'm not an engineer, but I can supply some capital and some programming help if some enterprising and learned person wants to point me in the right direction.

ryepdx: thank you. I don't have much of anything to contribute myself except a few wild ideas but I may buy some shares if I like the proposal. It could happen on Nafario's http://glbse.com/

It could be something that allows people to diversify their investments and actually trade the shares which may be rising even at times when bitcoins are flat or in a correction. Then they can hold, use or sell their dividends i.e. the bitcoins.

[N12]

GLBSE is a good idea, I’d probably buy shares.

[ryepdx]

Do you think it would be worth it? If you're willing to do this, then probably a few others are too. If you buy a rig like this for US$50000 ...

$50k is WAY too much for a 10Mh/s rig, IMO. Just right for a 30Mh/s to 50+Mh/s rig, though.

Maybe smaller investments is the way to go for bitcoins (unless you already have some fancy hardware). Buy something, generate a couple of blocks on it, then buy some new hardware that is better, and generate a few blocks on that, etc... Heavy investments will probably be overtaken by bitcoin limitations and technology before they've paid off.

Not if one gets in early enough. If an investment is to pay off, it doesn't matter if you've invested $2k or $20k. The number of days it'll take to break even stays the same. It's just a matter of how much risk you are willing to take on.

If so, what would be the right level of investments? If someone works on it, I guess it should be possible to find an optimal price level vs. block performance ratio that could be a buyers guide to get into bitcoins.

That would be interesting...

[Anonymous]

I like the idea of this. 
Now what can we do to make it happen ?

[caston]

Well I would suggest we make a list of the steps that companies normally go through for an IPO and take out all the boring and expensive regulation compliance. There would need to be a sort of prospectus and a business plan and cash flow forecasting. There would need to be break even analysis and profit and loss statements. These could be done in wiki form there would need to be other software as well possibly google docs like spreadsheets or modified versions of SMB Ledger.

Despite all the hardwork it would be a very exciting thing to be involved in.

You would deploy the miners in multiple locations around the globe so there would be reduced risk of natural disaster bringing down the company and you would make a special mining pool for these miners and backup servers in different locations for the mining pool server as well. You could even make it more p2p in nature so that any miner could be upgraded to become a pool server if need be.

You should also plan ahead so that more shares could be sold to raise more capital for further expansion and upgrades for example things even better than FPGAs http://www.technologyreview.com/computing/37406/?p1=A1&a=f. The idea being that the open and public FPGA miner grows bigger and spreads far more tentacles than any secret and private FPGA miner.

To stop the company from getting over a 50% share everything is open source so that others could make a similar company.

[greenlander]

Well, I have to point out that the $100 cost I pointed out is just for the FPGAs themselves.

You've got to have someone draw up board schematics, lay out a board, fabricate it, possibly spin it if there are hardware design bugs, assemble the prototypes, debug it.  You still need some kind of host system for it.

By the time you do all that you're looking at probably $20000 instead of $10000 for a 10 Gigahash system.  Some of it could be made up in volume by building 10 or more such systems... but some of it can't.  Assembly/fabrication costs are always going to be large for small quantities.  So now you're looking at a $/hash that's very close to PCs with ATI cards.

You'll also need time to get this running.  If you started today and did it right, you could be running in a couple of months... at which point the bitcoin difficulty will be quite a bit higher.

[colossus]

If T hash rigs start to appear it will send GPU mining the way of CPU mining and essentially destroy a little of the spirit of bitcoin, at the moment its open to all and the reward for participating is actually a very good incentive to participate even if you have just 1 card.

 Large FPGA clusters basically as stated skew the shift of power so much that people will not participate and the reward goes to the few as opposed to the many,

Also in my opinion this idea of a uber mining company is exactly what bitcoin wanted to avoid that not one organization controls the network or the flow of money, you become shareholders essentially in a bank and your proposal is a world bitcoin bank!

PS where do i signup .... here are my 30 silver bitcoins.

[ryepdx]

By the time you do all that you're looking at probably $20000 instead of $10000 for a 10 Gigahash system.  Some of it could be made up in volume by building 10 or more such systems... but some of it can't.  Assembly/fabrication costs are always going to be large for small quantities.  So now you're looking at a $/hash that's very close to PCs with ATI cards.

That's still within the range of what I'd be willing to pay for a 10Gh/s machine, especially if the machine had the same footprint as a single tower. I would prefer this over 10Gh/s with GPU machines due to the space savings.

If T hash rigs start to appear it will send GPU mining the way of CPU mining and essentially destroy a little of the spirit of bitcoin, at the moment its open to all and the reward for participating is actually a very good incentive to participate even if you have just 1 card.

How is this different from when GPUs sent CPUs the way of the dinosaur? If FPGA rigs remain modular, such that one could buy a 1Gh/s pluggable card for around $1,500 to $2,000, there really wouldn't be any difference from the situation we have now. Except this time the money would remain in the Bitcoin community rather than going to ATI or NVidia.

At any rate, I imagine the rise of mining corporations, or at best cooperatives, is inevitable. The lone hunter gives way to society. At best we can prepare for when it happens.

[colossus]

How is this different from when GPUs sent CPUs the way of the dinosaur? If FPGA rigs remain modular, such that one could buy a 1Gh/s pluggable card for around $1,500 to $2,000, there really wouldn't be any difference from the situation we have now. Except this time the money would remain in the Bitcoin community rather than going to ATI or NVidia.

At any rate, I imagine the rise of mining corporations, or at best cooperatives, is inevitable. The lone hunter gives way to society. At best we can prepare for when it happens.

Nice counter argument, i think the major difference is that CPUS/GPUS are common place, the FPGA is a bit specialised and intially it will be limited to those with the technical knowledge and the capital to go down that road.  Where as most can stretch to purchasing a gpu with a fair mhash rate.

I hear what your saying, we all knew FPGAs are out there, i guess i just feel the the short golden age of community powered bitcoin will drift away and it will be a battle of the corporations opposed to a battle between individuals.

[Sukrim]

Not a lot of "corporations" will be interested to invest in a market that small, volatile and unpredictable as bitcoin...

Most likely FPGA software will be released, a few people jump on it, drive difficulty to heights that are not fine for GPU miners anymore and the whole system collapses. As soon as some try to get back on board, someone just flips the switch on an FPGA cluster and it's over again.

People will lose interest in mining and - to be honest - it seems like the single most attractive thing about Bitcoin IS mining, not having any other benefit of a cryptocurrency. Even miners here seem rather to cash out at any(!) rate than holding onto their earnings for some time.

[neptop]

I don't think so. It's not too different from the GPU stuff I guess. The same fears. As long are enough people doing FPGA stuff there won't be a crash or something. I also think sooner or later someone will publish an open source version. Of course there will be people doing it from early on, but in order to not have the prices go down and to prevent people from losing interest one simply needs competition.

[colossus]

I don't think so. It's not too different from the GPU stuff I guess. The same fears. As long are enough people doing FPGA stuff there won't be a crash or something. I also think sooner or later someone will publish an open source version. Of course there will be people doing it from early on, but in order to not have the prices go down and to prevent people from losing interest one simply needs competition.

yeah but everyone had a gpu to start with  and it means corporations like the one proposed in this thread or co-ops perhaps for a lack of a better word, would control the bitcoin network.

Well i guess we see what happens.

[caston]

There may be some options. I am just getting ideas here but there is a company that offers low cost FPGA's: http://www.xess.com/

It appears that the older X1800 ATI cards used an FPGA to enable crossfire. You may be able to hack together a configuration where some of the workload from your late model card is offloaded to FPGA running a custom ROM on the X1800 ATI card. These cards seem to be listed for about $30 on ebay. No idea if this will work or even be worthwhile but it might be a step in the right direction.

http://techreport.com/articles.x/9138/2

There are more recent cards with like the PICO EX-160  that could also be used and potentially in combination with a GPU:

http://www.siliconwolves.net/frames/projects.html


There is also an intel CPU with integrated FPGA the Atom E600C series:
http://www.thinq.co.uk/2010/11/22/intel-launches-fpga-equipped-atom/

http://www.mardhikaputra.com/intel-introduces-new-processors-that-can-atom-e600c-configurable.html

And then there's this:

http://www.xtremedata.com/products/accelerators/pci-express/xdpcie3000

They also have some in socket (co-processor) accelerators but they are for older sockets.