I guess my argument is that the way you make your point about fpga being more of a hobby and GPUs being profitable, simply doesn’t work out in large scale.

If your doing this in your home, and don’t have to pay for space, time, maintenance, power consumption, and cooling costs. (for example in a proper datacenter setup) then GPU mining makes TONS of sense. In that setup you can re-use your GPUs as you suggested, and it makes a great investment because the GPUs can be resold for a reasonable price afterwards.

But if your planning on doing a large scale mining operation (as a business). And you consider the purchase price, constant upkeep costs, and so on. The GPU Solution is not profitable on a very large scale. (in otherwords at $5 per BTC and current difficulty +20% GPU mining would be loosing money each month, but the FPGA miner would be earning money each month. At that point it doesn’t matter if the FPGA costs 10 times as much, that point is irrelevant.)

So I guess what I’m saying is your point is valid but backwards… GPUs are viable for smaller scale home based setups (in the long run) where FPGA setups are more viable for large scale “professional” mining setups.

To back some of this up, here are some of my calculations (it’s hard to visualize without the VERY large spreadsheet I used to put this together, but suffice it to say I have put extensive planning into this, and plan on making a real business case around launching this on a large scale).

Using “best case estimates” for difficulty at 1,500,000 and BTC price in USD at $5…

If I max out a 3U Rackmount enclosure with high end cooling, I can cram 4x 6990 into a single enclosure. Along with a motherboard to handle it and so on. Using estimates of price/power from the mining hardware comparison site, (and assuming 10% savings for volume orders on the GPUs) that puts my total enclosure cost (4x GPU, motherboard, chasis, PSU, and all other components for one self contained unit) at about $3400 per enclosure by my calculations.

I can fit 12 of those in a full rack, accounting for switching infrastructure, and a pair of 1U servers acting as a load balanced pair of private pooling servers (so I have redundant pool which is hitting the bitcoin network directly).

Using the GPU option my total purchase/setup cost of a full 42U rack comes out to be $49,000 USD. Including the rack itself, and all supporting infrastructure to make that 42U rack a completely self contained mining operation.

That GPU rack would be capable of pushing out approximately 36GHash/s of mining power. which would produce approximately .48 blocks per day at the estimated difficulty I mention above. (estimated average, this would fluctuate based on luck of course, but should average out).

The total power consumption of the rack would be 22.6 KW including switching, and the pool servers.

If you assume the purchase price is financed, at 4% interest, and your target is for ROI in one year, your monthly “dept repayment” cost is $4,296 approx per month. Then factor in hosting/maintenance/labour/cooling costs of about $2000 a month for one full rack, and electricity costs (at least at the going rate here in canada) of $2,440 per month based on the consumption figures above.

Your total monthly operating costs would be $8,737 per month for the first year (after that you make $4,296 more in profits each month because you’re not repaying debt).

now at a rate of half a block per day, and 50 BTC per block (which will drop near the end of 2012) your looking at 25BTC per day (or more specifically 725.80 BTC per month). at $5 per BTC your loosing money every month (you would never repay your debt). Even after the dept repayment (if you assume you had zero debt to repay at all and the hardware was free, you still loose $811 per month). By my calculations with all other variables above as-is, you would need to have a BTC value of over $12 average for you to make any money at all.

Now if you spread the debt out over 10 years, you can reduce that “minimum value to break even” down to $7 per BTC.

Now consider the same rack enclosure, using currently available FPGA solutions:

With a single FPGA board from ztex, I can fit about 64 boards in a 2U rackmount chasis with good cooling. This can use cheaper motherboards for control/software because it’s all USB interface (using hubs). and the software does no real work. So I can use lightweight low power single board PCs as a “controller” for that enclosure. Bought in bulk these boards can be had for about $180 each. That makes for a total enclosure cost (including all other components to build a standalone mining enclosure) at about $12,560.

I can fit 18 of these enclosures in a 42U rack with the same supporting infrastructure (switching/pooling and so on) as the GPU option.

So for the FPGA option the final full rack setup costs would be $234,000 for the exact same footprint as the GPU option. including all FPGA boards, enclosures, infrastructure, and the rack, and so on to make it a completely self contained mining operation.

That FPGA rack would be capable of pushing out approximately 230GHash/s of mining power which would produce approximately 3.09 blocks per day on average. (assuming all the same difficulty specs as the gpu option).

The total power consumption of a full rack would be 12.93KW including all infrastructure and such.

Assuming all the same financing options on the purchase price as the GPU option (4% rate) for the total up front costs and startup costs, and the same ROI (1year) you would be looking at a monthly debt repayment cost of $20,477, with a monthly hosting/maintenance/labour cost of $2000 a month for the cabinet, and electricity costs of $1,397.

Your total monthly operating costs would be $23,875 for the first year, after which you would be turning $23,875 more in profit each month because the debt is repaid.

At the above mining rate, assuming $5 per BTC on average it would take just slightly longer than 12 months to pay off your initial investment and all interest in full (the calculation shows it would require an average BTC/USD value of $5.14 to pay it all off in full after exactly 12 months).

And in the long run assuming a $5 average BTC rate, as compared above your making a total of about $20K per month.

You don’t need to spread the debt any further at all…

I also considered designing my own custom FPGA solution based on the opensource software from ztex, and moving to a higher density board, to improve density and volume purchasing costs (reducing the per-board cost). Doing that your looking at putting 8x FPGAs on a single board. even calculating in additional up front costs of $40,000 USD in development costs, and board design costs, plus the increased purchase cost per board (but the updated performance as well) you would be looking at an overall payoff in only 8 months at $5 average BTC price using the “custom” boards, and a monthly profit after ROI from the same 42U rack (but using the 8x boards) of $41K per month.

So looking at these numbers above, I’m not sure how you can possibly think that FPGA mining is not profitable compared to GPU mining. At least in a professional large scale mining operation, it’s the exact opposite.

Then as I pointed out, consider that the next generation of ATI cards are very likely to be much SLOWER at mining than the current generation cards, that doesn’t fare well for the future proofing of a GPU based setup.

Also to counter the resale value, with GPUs you can’t expect to make more than 50% of your investment back on resale. (drop in value of the cards over time, and the fact they are used).

FPGAs should be able to get close to that return as well (since your dealing with a massive scale cluster, they can be repurposed for scientific use, security uses, or broken up and sold as smaller units for FPGA development). The longterm price of FPGAs is much more stable than that of commodity PC hardware, but because it’s a more niche application I suspect resale would be lower by a bit. Which is why I expect approximately a 50% resale value within the first 2 years of their life (beyond that who cares, you’ve just mined a quarter million in pure profit in a year and the hardware is fully paid for, so who cares if you can resell it, that’s just icing on the cake).

Thoughts?

One of the points I raised in my article is that the FPGA devices are not multi-taskers, in that if I buy an FPGA device for ~$500, all it’s doing is mining bitcoins. For by that logic, you’re buying the device to accumulate Bitcoins. Your point that over time, difficulty will go up also means that (more than likely), price is going up.

So as a consumer with $500 burning a hole in my pockets, I can make the decision to either:

1) Buy Bitcoins with my $500 and speculate that because difficulty/price is going up, this is a wise investment. (~96 Bitcoins)
2) Buy a FPGA device and let it mine Bitcoins for me (96 coins at ~300mhash? ~13 months + power)
3) Build a mining PC and let it mine Bitcoins/Game (96 coins at ~40mhash? ~10 months + power)

The point I’d like to add to my original point is that because FPGA solutions aren’t multi-taskers, you’re essentially making the decision between Option #1 and #2 above since the device cannot be used for anything else.

So I debate the point people make on FPGA profitability because it only focuses on mining/power consumption without external factors, such as start up cost, some areas do have lower power rates (or even free for some folks) and resale value of the hardware.

Going FPGA is going “All In” on Bitcoin. I’m not willing to take that kind of risk and building traditional GPU based mining rigs gives me a good level of income (low start up costs, including energy costs) and a strategy for getting out by selling my mining rigs as “Gaming PC’s”, which would go toward recovering the start up cost of a final pulse of income.

I welcome anyone thinking of getting into Bitcoin to check out the charts available on http://bitcoinstatus.rowit.co.uk/ and make your own decision.

Anyway the big thing that was overlooked in this article, is that FPGA as a solution isn’t flawed, the state of the designs for FPGA solutions was young when this was written. But with the same hardware, and advances in the FPGA designs, we’re now seeing FPGAs that can be had for $180 in volume (full board solutions) which can pull 200MHash+ and draw only 8W of power…

I’ve done the math on those solutions versus a GPU solution, and especially in areas with higher electricity costs, and with the lower value of the bitcoin right now. And considering some operating costs due to size, cooling, and electricity requirements. A large scale operation using FPGAs, versus the same solution using GPUs (assuming the highest achievable density for a standard 19″ 42U Datacenter Rack) the GPU solution would never pay for itself, and the FPGA solution would recoup 100% of it’s investment within 10-12 months. (and make a killing after that).

Also, consider that right now MOST GPUs can’t mine worth of crap. Because the architecture is not optimized for highly parallel integer operations, which are what is needed for bitcoin hashing. Only the ATI cards out now are optimized for this. Unfortunately ATI is moving towards the other way of doing things in it’s next series of cards. Which will likely result in performance actually DROPPING on GPUs in the next round. Making them far less efficient.

Also I think you too easily dismissed the problem of shifting difficulty/value ratio making GPUs less profitable due to electricity costs. As it takes more MHash to earn USD $ for example, with a fixed electricity costs in USD, eventually you will hit a break even point, where your GPU stops making money, because it only makes enough to barely pay it’s electricity. At that point, an FPGA that uses 8 watts for 200MHash, versus a GPU using 350Watt for 400MHash, (round numbers off the top of my head). Your talking about 0.875 MHash/s per watt for the GPU solution, and 25MHash/s per watt for the FPGA… that puts it at at least 20 times as power efficient. Which means that when that GPU hits it’s “break even” point. That FPGA is still churning away at a 95% profit margin (on operating costs). So at that point the GPU will never pay for itself, but the FPGA will EVENTUALLY pay for itself (regardless of differences in purchase cost).

I do realize that when the article was written, the FPGA was about 50% performance of what it is now (The same FPGA that does 200+Mhash/s now did 100MHash/s then). but even at those rates, the power efficiency would have become a problem for the GPU option in the near future making FPGA more profitable.

Just thought I’d post this update here, so those finding this article via google don’t get an incorrect view of the value of FPGA mining (which is definitely NOT just a fad/hobby and is actually potentially MUCH more profitable than GPU mining right now)

Competition is the key to success, and the easiest way to assure limited competition, and by extension assure currency failure, is to accept any semblance of “legal tender” on the web. Perhaps BitCoin has a place, but it is in a more diverse ecosystem where multiple credit clearinghouses operate and compete for transaction fees.

I’m working on doing a WCF service for a big corporation and was wondering if I could take a look at your code to see if I could implement some of your tricks into my code.

Thanks!

Angie

It’s one of those things I really hope we’re working towards, where you have “Game Profiles” for your video card, we’ll also have profiles for the CPU where you can tell the system to overclock the crap out of one core and park the others since the game is single threaded.

My hope is based on AMD/ATI being the same company, so hopefully the idea will cross hardware lines.