Topic: BTCMiner - Open Source Bitcoin Miner for ZTEX FPGA Boards, 215 MH/s on LX150 - page 31. (Read 161816 times)

The Easypath program is only available for Virtex FPGA's and the cost reduction is only 35%. This is still more expensive than Spartan 6 FPGA's.

The frequency limit is about 135 MHz (non overclocked) with the pipelines I used. This limit is defined by the levels of logic. Higher frequencies are only possible with longer pipelines. But these pipelines are not routable on LX150 anymore.

Unfortunately LX150's (at least the ones I have) can't be overclocked much.

Agreed. Somebody might make a toy full-custom ASIC on a low-cost academic fab (~250nm), but a modern FPGA will beat it on performance per watt and per dollar. A 40nm mask set costs $4M -- I don't think anyone will invest that kind of money into Bitcoin at the moment.

Xilinx Easypath could make sense once the HDL designs stop improving. But I think we will see 6s150 reach 175-200MH/s, OC'd, before then. And I'd hate to submit a final design into Easypath 1 week before somebody releases a 10% optimization to the RTL. That would suck.

-rph

The main problem about asics is the initial costs.
There are basically 2 kinds of "asics":

Cell based asics which are built like fpgas with the difference that the logic is fused roms and the interconnect is hard-wired. Those have the lowest initial cost but also the weakest performance. There are different many different products which offer different block layouts and interconnect strategies.
The main advantage is that these chips can be produced from HDL code.

Custom asics are a different animal. You work on bare silicon, from scratch. You have to work on the analogue domain here, write your own software to model your transistors, choose between doping, layer thickness, model interconnect, calculate resonances and so on.
From what I've heard every company writes its own proprietary software and uses that, there is FOSS toolkit or commercial one for that.
Even worse every software uses different models, and the more you can get to the cutting edge of physics the better the chip will be.

The research and the software alone are several man years. As for bitcoin... if it ever becomes as big as lets say Linux you might start thinking of an effort but right now this looks like a reach for the stars.

So if FPGA is "too versatile" what is keeping ASIC folding development from taking off?
Thanks for bringing up this point.

The problem is FPGAs themselves are very expensive and in order to get them to work you need a board communication / logic voltage level conversion and power converters.

So while being the most versatile silicon estate they are also the most expensive. The also have alot of i/o capabilities (the most of any ic) which we don't need for bitcoin mining but have to pay nevertheless.

How do you define "efficiency"? Usually efficiency means H/J. LX150 FPGA boards generate about 10 times more H/J than modern 6990 GPU's. This ratio is constant over the time.

Let's consider that we are keeping our rigs for the long haul and Not reselling the GPU (which is a great point btw). Say we are building these machines to not only mine Btc but any and all bitcoin forks and folding operations as need be.
I build two identical machines except in one I have a high end GPU and the latest FPGA (lets say X10 in order to hash the same amount per second.)
If I'm understanding this right, the GPU looses its worth because of the energy it takes to make a Btc becomes higher then the profit from the usd conversion (at current prices) due to increasing difficulty and decreasing reward (50 to 25 etc).

For my inquiry lets pretend I have a Time Dollar system set up and use a crypto currency like BitCoin to keep track of who has what. The BTD (bit time dollar if you will) always has a constant worth of one man hour and the difficulty will remain the same throughout the life time of the project (although the reward significantly lower.)

Keeping in mind this project is for closed communities and although there is still an incentive to Mine coins the major incentive is to be part of the project and not necessarily become resource rich off the process because they have no $ worth (maybe but not for the scope of this project)
This rig is to be built as the main / first mining server pool until the others get involved so having better then everyone else is not the goal.

So my question in this case would be how long until my FPGA machine becomes more cost efficient then the GPU Rig.  I can totally see the possibilities in cost efficiency if setting up home miners with just one card for the casual participant But when setting up the "backbone" which would be the better investment for the community.
Thanks for all your input.

@ztex: I just want to ask, when we order the FPGA boards, would they be just "plug & play" / ready to run? Or will I need other parts? Basically what are the complete list of things I need to make this FPGA board run?

I know I sound like I don't know what I'm talking about but the uber low energy cost of operation is very enticing compared to GPU mining. It has been a great learning experience BTC has been. I hope you could answer my queries.

Another factor to consider is depreciation / resale value. You can mine with a GPU for 6 months and then sell it for nearly what you paid on eBay. FPGAs have much less of a secondary market.

Still, my money is on FPGAs having a better long term ROI, as they will be profitable operationally, as long as the difficulty factor is defined by GPUs.

-rph

Forget the values from the table of my previous post. They are obviously wrong (due to a typo in the formula I entered in the spread sheet program) : If the profitability (revenues in USD per day) is halved per year the reference 6990 GPU will become unprofitable after about one year. This is not reflected in the table. (No critical readers here?    )

With the correct calculation only the 100 board cluster would get the the money in and all LX150 FPGA boards would become unprofitable after about 4.3 years. In that case no one would invest in new hardware. But if no one invests in new hardware the profitability is not reduced -- at least not by a factor of 0.5/a ...

Obviously the model I used in the previous post is too simple. A more realistic scenario is that the profitability remains almost constant for a while (until the oldtimers are squeezed out which happens now). But at latest after the the BTC/block reduction end of 2012 the profitability is halved. After that event the profitability is reduced slowly until current GPU's are squeezed out either by FPGAs/ASIC's ore more efficient GPU's.

The following calculations are based on profitability reduction of 0.5/a for a little bit more than a year (until the 6990s become unprofitable) and than a profitability reduction by 0.8/a.
This is a bad case scenario and more pessimistic than the scenario described above.

		price in	current revenues	energy costs	revenues in	revenues in	revenues in	revenues in
rig	GH/s	USD	in USD/d	in USD/d	USD after 378d	USD after 820d	USD after 1650d	USD after 2400
1 Board	0.135	471	0.65	0.03	163.97	273.52	406.15	471.29
25-cluster	3.375	10000	16.23	0.82	4067.47	6771.46	10024.14	11597.58
100-cluster	13.5	27000	64.92	3.36	16239.64	27020.23	39964.55	46198.34
6990x2	1.436	1800	6.91	3.37	589.85	405.86	-716.63	-2312.39

Indeed thank you Immensely. 

So roughly 10 Boards to equal two 6990s for GH/s

10*$500=$5,000.00
$720*2=$1,440.00
$5,000-$1,440=$3,560.00
So as of now it boils down to how much electricity the Boards will save and the ROI. How many years worth of electricity you could have purchased with that. Well till the $ comes down a bit....

I have much faith in the futures of this hardware. Especially for those enterprising bitminers seeking to use alt energies to run their rigs. It is much more feasible to run a solar rig now thanks to these.
Thank you for your work on this technology.

I just want to say thank you for taking the time to write up that post. There's a lot of back'n'forth currently on whether we're seeing a shift from GPU to FPGA (and ASIC) at all, sooner or later at the moment. More data always help.

(More calculations in this thread, but also a fair bit of harsh language: https://bitcointalk.org/index.php?topic=40865.20 )

Please read the initial posts, there will be a cost optimized version (expected for end of September) and there are volume discounts.

Profitability calculations
Here are a some profitability calculations. The 25 cluster price bases on the 25 unit price of the cost optimized 1.15x version (as stated in the initial posting). The 100 cluster price bases on the costs if the boards are produced under license.  Both cluster prices contain approximate costs for a small PC (Atom is sufficient) + hubs+ cables + fans + self made case. The GPU prices are taken from https://en.bitcoin.it/wiki/Mining_hardware_comparison (I cannot find a "X6500" GPU's there)

The calculation bases on the current difficulty, 8.50 USD/BTC, 0.20 USD/kWh and a profitabilty reduction by a factor of 2 per year. This scenario is realistic if the bitcoin mining market is saturated (if difficulty becomes stable as now).

		price in	current revenues	energy costs	revenues in	revenues in	revenues in	revenues in
rig	GH/s	USD	in USD/d	in USD/d	USD after 500d	USD after 570d	USD after 640d	USD after 710
1 Board	0.135	471	0.65	0.03	272.45	336.34	402.96	471.61
25-cluster	3.375	10000	16.23	0.82	6767.52	8357.79	10016.09	11725.37
100-cluster	13.5	27000	64.92	3.36	27030.09	33385.54	40013.16	46844.67
6990x2	1.436	1800	6.91	3.37	1370.86	1836.46	2331.03	2847.31

50BTC/block -> 25BTC/block
The BTC/block reduction at end of 2012 will squeeze out many GPU miners. IMHO, after the event the difficulty will be reduced until most efficient GPU's are just profitable. (But who knows, if there will be many FPGA (or even ASIC) miners only they will survive.)

EDIT -- forget the values from the table, see https://bitcointalksearch.org/topic/m.505088

Help me out here, How many USB-FPGA Module 1.15d (XC6SLX150, SG 3) $496.21 would I need to make the hashing power of say the AMD Radeon™ HD 6990 for around $720? And if you could sum up the difference saved in energy showing how the FPGAs are more efficient.

Also; How does this FPGA stand against the Dual FPGA X6500 $610

I am only assuming that this equipment can also fold for organizations like SETI and not just BitCoin specific. Cause even though the ฿TC may come and go the crypto currency genie has been released.

My main project is to put together a system for local communities to run their own time dollar system using BitCoin as the book keeper.  One of the main components would be a main server to kick start the system as other members start to connect etc.  And If I were to put a package together It would need to be the most bang for the ฿TC so to speak.

Thanks for your time.

I appreciate the portability though.

Would be a bit nicer if it didn't use java. At least it's not TCL though. 

Nice work Zetex!   

done

I'm sorry, but I did not receive your request. I even checked the mail logs for failed attempts.

Please either use the contact form at http://shop.ztex.de/contact_us.php or the email address stated in the imprint http://shop.ztex.de/imprint.php