How to determine if an FPGA is suitable for mining?

[Wandering Albatross]

I've looked at some of the FPGA miners available and am curious to know how to determine if an FPGA is well suited to mining.
What characteristics make an FPGA suitable or not?

[1713437263]

1713437263

[1713437263]

1713437263

[1713437263]

1713437263

[1713437263]

1713437263

[1713437263]

1713437263

[Stephen Gornick]

I've looked at some of the FPGA miners available and am curious to know how to determine if an FPGA is well suited to mining.
What characteristics make an FPGA suitable or not?

There's some performance info here:
 - http://en.bitcoin.it/wiki/Mining_Hardware_Comparison#FPGA_Devices
 - http://bitcoinfpga.com

[qualalol]

I'm no expert, but as far as I can tell the FPGA itself is less important than the circuit that is designed to run on it. (You can see a list of common boards under https://en.bitcoin.it/wiki/Mining_hardware_comparison#FPGA_Devices ). The characteristics seem to be lowest cost for a high number of blocks, as well as clock frequency... Since bitcoin doesn't need much I/O that factor can probably be ignored, and the power use is small enough not to play a major role in selection.

[Wandering Albatross]

Yep, thanks, have read those links, I will presume that the folks putting together the FPGA miners have found the optimal solution. But I am curious how far and wide they have looked.  There must be some novel approaches to still be considered as well.

[heybabit]

Has anyone actually constructed one of these with cheaper parts just to mess with this idea? I like attempting to put things together myself but the high end FPGAs require special soldering etc. Maybe someone has done a simple (inefficient) low end project...

[Wandering Albatross]

Has anyone actually constructed one of these with cheaper parts just to mess with this idea?

Cheaper than what? What parts would you use? Can the problem be reduced to simpler hardware? Is that your question?


On one page they describe a project that costs $175.  But I think they all use a spartan chip.

I too was wondering if the problem can be solved by 200 $0.50 chips or something like that.

I haven't seen a page that shows the hardware depreciation also as a function of POW difficulty increase over time.
I will have to ask the taxman if I can start a business, buy mining hardware and use conventional depreciation as allowed currently (U.S.) to write-off the hardware costs and also increase the rate of depreciation due to POW difficulty increasing.

If difficulty doubles than that hardware does half the work all of the sudden and the investment goes into the toilet.

[CA Coins]

The most common chip used by most of the available FPGA boards is the spartan 6 LX150.  You can try to assemble a board yourself, Icarus is open source, Ztex will sell you the board and BOM, but the BGA is difficult to solder on and you will probably want to hire an assembler.  You might save a few bucks, but if you get them in bulk, the price is not too bad.  I think the Icarus is $469 in quantity of 30.  The spartan 6 will do about 200MH, give and take a bit with the different boards/software.  I think butterflylabs is advertising a different FPGA/ASIC, but I don't think it is in production yet.

EDIT:  Oh, in the US, your depreciation schedule is dependent on the type of equipment.  Electronics I think is 5 years by the straight-line method.  I would be very surprised if the IRS will account for changes in difficulty for the depreciation, but please let me know what your CPA says.  You can also take a section 179 to write off all of it in a year if you qualify. 

[Wandering Albatross]

butterflylabs is advertising a different FPGA/ASIC, but I don't think it is in production yet.

Sounds interesting will have to look for that.

EDIT:  Oh, in the US, your depreciation schedule is dependent on the type of equipment.  Electronics I think is 5 years by the straight-line method.  I would be very surprised if the IRS will account for changes in difficulty for the depreciation, but please let me know what your CPA says.  You can also take a section 179 to write off all of it in a year if you qualify. 

Don't do my own taxes now but I thought it was shorter than 5 years.  There may be some creative searching needed to find the correct characterization of the mining hardware losses due to increase in difficulty.  I'm trying to think of an analog in another field. Hmm, the IRS pays mileage and part of that is wear/tear. It may require a special case. A craftsman can depreciate tools that are required for their trade.
They key issue for bitcoin miner hardware is how fast and by how much can the HW be depreciated. It would have to be coupled to increase in difficulty of course.

[DeathAndTaxes]

Don't do my own taxes now but I thought it was shorter than 5 years.  There may be some creative searching needed to find the correct characterization of the mining hardware losses due to increase in difficulty.  I'm trying to think of an analog in another field. Hmm, the IRS pays mileage and part of that is wear/tear. It may require a special case. A craftsman can depreciate tools that are required for their trade.
They key issue for bitcoin miner hardware is how fast and by how much can the HW be depreciated. It would have to be coupled to increase in difficulty of course.

You seem to have difficulty listening.  It was already answered.

EDIT:  Oh, in the US, your depreciation schedule is dependent on the type of equipment.  Electronics I think is 5 years by the straight-line method.  I would be very surprised if the IRS will account for changes in difficulty for the depreciation, but please let me know what your CPA says.  You can also take a section 179 to write off all of it in a year if you qualify. 

The IRS doesn't give a flying shit about difficulty going up.  Depreciation for tax purposes has little correlation between depreciation in the real world.  In your craftsman example if the craftsman buys crappy tools and 3 months later realized they are crap, sells them and buys new ones the IRS isn't going to allow a 6 month depreciation.

[Wandering Albatross]

nothing worth quoting

Ignore.

[heybabit]

Has anyone actually constructed one of these with cheaper parts just to mess with this idea?

Cheaper than what? What parts would you use? Can the problem be reduced to simpler hardware? Is that your question?

Sorry I should have been more clear in my question. I'm not looking necessarily for the most efficient parts for mining. I'm looking for a more simple construction to play with and get familiar with the process, FPGAs and software etc. before sinking more money into more expensive parts. If no one has really done this or there really isn't a cheaper entry point just to mess around, then I'm ok with that.

I have experience with building simple boards but nothing with complex mounting like the Spartan 6 LX150 or others.

[Wandering Albatross]

If no one has really done this or there really isn't a cheaper entry point just to mess around, then I'm ok with that.

I have experience with building simple boards but nothing with complex mounting like the Spartan 6 LX150 or others.

Define cheap? I'm cheap too and it would be fun to play with something that was $50 or so. That way if I destroy it I don't care that much. That's why I was asking about how to determine if a given FPGA would be suitable for doing SHA256. But maybe there are other cheap devices that could be put into service for doing the SHA256.

[heybabit]

Define cheap? I'm cheap too and it would be fun to play with something that was $50 or so. That way if I destroy it I don't care that much. That's why I was asking about how to determine if a given FPGA would be suitable for doing SHA256. But maybe there are other cheap devices that could be put into service for doing the SHA256.

Now we're thinking along the same lines. $50 would be cheap enough. I just haven't done enough research to see if there is anything in that range. I've been interested in FPGAs for other reasons as well, just never jumped into it yet. If there is something in that range, I'd be willing to give it a try.

[RandyFolds]

nothing worth quoting

Ignore.

Well, that was probably a mistake. Don't ignore the one educated individual who is actually giving sound advice.

The work involved in getting cheaper/less powerful chips to process a bitstream in tandem is going to be ridiculous. I don't know a damn thing about it, but my man that you've chosen to ignore sure does.

[Wandering Albatross]

Now we're thinking along the same lines. $50 would be cheap enough. I just haven't done enough research to see if there is anything in that range. I've been interested in FPGAs for other reasons as well, just never jumped into it yet. If there is something in that range, I'd be willing to give it a try.

There's a guy on these forums trying to get pre-orders for an FPGA rig based on an Altera. The devkit is $86.
I may try it but not sure I want to wait until July. His project isn't starting until he's funded. That chip was announced in 2009 but appears
to be lower power than spartan6 (which other fpga miners are using).

[Davyd05]

FPGA, keep me on the fence about starting a mining rig.

[pieppiep]

I found a very cheap fpga development board at http://www.dhgate.com/altera-fpga-cycloneii-ep2c5t144-minimum-system/p-ff80808133cfd7010133d0a6374a7d0f.html
Only $35.88 for one and even less if you buy multiple.
I'm sure it would be possible to do some bitcoin mining with it, but the speed would probably very low.
If I can get 1MH/s the thing can mine, but would be about 10 times more expensive than for example the icarus.

[DeathAndTaxes]

That board is useless for mining.

An unrolled double SHA-256 loop requires ~ 120K LUTs.  Remember Bitcoin is a double hash.  So for example A Spartan -6 LX 150 has 150K LUTS so it firts 1 complete unrolled loop. Every clock cycle it completes 1 hash.  Running at 200 Mhz it performs 200 MH/s.

The chip you linked to has 5K LUTS.  It is impossible to fit an unrolled SHA-256 loop.  You could look into using a looped version (where each clock the chip performs 1 round of SHA-256 process) but performance is going to be very low.  There is significant overhead in handling the loop combined with the need to save values intra-rounds means efficiency (hashes per LUT) will be bad.  The chip is also 2 generations old meaning clock speed is going to be much lower.

Ballpark that chip might be able (with a huge amount of work customizing a bitstream) get 0.1 MH/s.

[Wandering Albatross]

altera-fpga-cycloneii

Latest is CycloneV
You want to future-proof whenever possible. But I'm not convinced the Cyclone is the right HW.

But really the key is free power. If you can vampire power for free then you are ahead.

I wanna see a TeslaMiner. There's free energy everywhere!
I predict Iceland will use geothermal to mine a huge portion of BTC. (if they're not already)

So the ultimate rig would be one that costs nothing to power it.  FPGA or GPU wouldn't matter.
Maybe BTC will be beneficial to society in that it will force people to improve novel ways to mine free energy.
(or at least freer energy than 0.03 kWh)

[us11csalyer]

Call it unethical but my friend from college has a 4x5970 setup running 24/7 in his office at work. He has a few more of the same rigs running 24/7 at his sister's house because she is on some low income energy program where she pays a flat rate no matter the usages.

Tapping power that is being paid for by the state/government doesn't matter to me. We are paying for it one way or another.