Standalone FPGA miner board (large scale)

[pdki]

Hi,

I was following the thread "Official Open Source FPGA Bitcoin Miner (Smaller Devices Now Supported!)". Unfortunately I am a newby and cannot post over there, now I do it here:

I have found a standalone FPGA board with ARM linux system incl. network that is probably not so far from being competetive in the MH/$ region:
http://www.dinigroup.com/new/DNDPB_S327.php

It has 27 Cyclone III with 120K LE. From the above mentioned thread you can get a full SHA256 pipeline into ~75K LE running at 100MHz = 100MH/s.
27*120K LE = 3240K LE => 4.3GH/s

Probably with less than 150W power consumption. VHDL code is available from the other thread, as its linux, one of the available miner cores should be easy to adapt.

Price: ~21K$ :-(

The FPGAs alone should be somewhere in the range of 8K$, so there is probably potential for a price reduction, if we buy enough of the boards.

With power consumption becoming more and more important (in Germany), anybody interested in buying some boards?

Of course an ASIC solution is far superior, but will require much more work and cost upfront.

[1714965633]

1714965633

[1714965633]

1714965633

[Grant]

Compared to other FPGA's i've seen this is a pretty good price.

But potential ROI ?

Assuming we get an average price of 10$/BTC and average difficulty of 4mill... Would yield ~$328.70/mo, or 63 months to payoff. (thats still ~18% return per year, if this holds which is pretty good as an investment)

[N4rk0]

Compared to other FPGA's i've seen this is a pretty good price.

But potential ROI ?

Assuming we get an average price of 10$/BTC and average difficulty of 4mill... Would yield ~$328.70/mo, or 63 months to payoff. (thats still ~18% return per year, if this holds which is pretty good as an investment)

The problem is that the difficulty is increasing, in the next inrease will be almost doubled and so also the payback time , and also you can't forsee what will happen in 63 months.

[itsmine]

I don't know about anyone else, but I would be interested in ajoint venture with the OP or someone else  that would consist off buying high grade hardware like op mentioned and hosting it in like Alaska.

If you get 10 people the price is greatly reduced.  And if we get is good enough may even make some money.  Sure it is a gamble, but a lesser gamble for all those involved.  Don't be greedy now

Alternatively we can have the chinese make a copy...

[nemo]

I used to buy broken plasma and lcd tvs and I've seen a couple fpgas on the controller boards. I have reball experience and equipment. Too bad I can't design and print pcb for shit, or program. On hackaday some guy took a lot of these and built an array for cracking hashes. It took him months. It's just a shame that tens of thousands of dollars if not hundreds of thousands (dozens of 40 foot trailers packed to the brim with tvs) of fpgas are sitting in an lcd/plasma graveyard about 30 miles from my house.

Here is the article if anybody is interested:
http://hackaday.com/2007/08/31/nsahome-diy-shared-fpga-cracker/

It doesn't keep me up at night or anything, it's just a fun thought to entertain.

[kjj]

Hi,

I was following the thread "Official Open Source FPGA Bitcoin Miner (Smaller Devices Now Supported!)". Unfortunately I am a newby and cannot post over there, now I do it here:

I have found a standalone FPGA board with ARM linux system incl. network that is probably not so far from being competetive in the MH/$ region:
http://www.dinigroup.com/new/DNDPB_S327.php

It has 27 Cyclone III with 120K LE. From the above mentioned thread you can get a full SHA256 pipeline into ~75K LE running at 100MHz = 100MH/s.
27*120K LE = 3240K LE => 4.3GH/s

Probably with less than 150W power consumption. VHDL code is available from the other thread, as its linux, one of the available miner cores should be easy to adapt.

Price: ~21K$ :-(

The FPGAs alone should be somewhere in the range of 8K$, so there is probably potential for a price reduction, if we buy enough of the boards.

With power consumption becoming more and more important (in Germany), anybody interested in buying some boards?

Of course an ASIC solution is far superior, but will require much more work and cost upfront.

Your math is off by a bit because you can't split cores.  If a core takes 75K LE, you don't get three of them by joining 2 chips with 120K LE each.  You get 2.

So, 27 chips would give you 2.7 Ghash/sec instead of 4.3 Ghash/sec.  I don't know how that factors into your projections.

[LazarusLong]

Anybody knows which HDTV sets these chips are inside?
  I could buy up all broken TVs on ebay...

[nemo]

Most plasma and lcds I've seen have one, sometimes two. The newer the better.

[makomk]

Your math is off by a bit because you can't split cores.  If a core takes 75K LE, you don't get three of them by joining 2 chips with 120K LE each.  You get 2.

So, 27 chips would give you 2.7 Ghash/sec instead of 4.3 Ghash/sec.  I don't know how that factors into your projections.

In theory it might be possible to fit another half-core in each chip doing a hash every 2 cycles and hit 4 Ghash/sec total, but it'd be far from trivial to achieve.