Ultra-Low-Cost DIY FPGA Miner - 175MH/s @ $1/MH

[rph]

Without releasing too many details - my FPGA cluster will pay for itself in under 10 months
at current price and difficulty, after power costs.

No GPU will ever pay for itself, with socal power costs, at current price/difficulty.

Basically: FPGAs have already won, unless you are getting free or almost-free power somehow.

-rph

[amazingrando]

Without releasing too many details - a 100 FPGA cluster will produce its own cost in about 8 months
at current price and difficulty, after power costs.

No GPU will ever pay for itself, with my (socal) power costs, at current price/difficulty.

Basically: FPGAs have already won, unless you are getting free or almost-free power somehow.

-rph

I'm also in Socal.  GPU mining hasn't been profitable here for many months, so I moved all of my equipment north 1,000 miles.  GPU's do give me the benefit of being able to easily sell the equipment when I decide to exit mining.   But I would easily invest $10-20k in FPGA's if they could hit the right price/performance ratio.

[rph]

I'm also in Socal.  GPU mining hasn't been profitable here for many months, so I moved all of my equipment north 1,000 miles.  GPU's do give me the benefit of being able to easily sell the equipment when I decide to exit mining.

GPUs do have a better, more liquid resale market than FPGAs, but you also need to upgrade them more often.
5830s will be totally useless for BTC mining, once 28nm GPUs in $0.05/kWhr areas are driving the difficulty.

Whereas, 45nm FPGAs will be profitable for 3+ years, as long as difficulty is driven by GPUs.
They're 8-10X more power efficient. So basically there is much less risk of them becoming unprofitable
before they pay for themselves. Set it up, forget about it, and cash the checks each month.

I think that as awareness grows, there will be a decent secondary market for FPGA miners.

-rph

[rph]

Updated the pics, since people were making fun of the tangled-wire prototype  

It's almost professional-looking now. And I got the full-size heatsinks.

The next step is a carrier that accepts a crapload of these FPGA modules.







-rph

[DeathAndTaxes]

Looks AWESOME.  I like how small it is and how you made a "FPGA socket".  Any plans in the future to make larger board that have multiple FPGA sockets?

What is the timeline for taking orders?  Are you still getting 175MH/s?  Estimated pricing?  Discounts for bulk orders?

I think 2012 might be the year for FPGA.

2009 - Bitcoin prototype, unoptomized miners, mostly solo mining
2010 - Rise of the pools and optimized miners.
2011 - GPU take over
2012 - Commercial FPGA miners make their entrance

I have no more power unless I have an electrician run a new 230V line for my garage.  10GH is a lot of load and heat.  I couldn't do another 10GH of GPU if I wanted to without getting some commercial space.  However 10GH of FPGA .... 

[rph]

Any plans in the future to make larger board that have multiple FPGA sockets?

Yes absolutely.

What is the timeline for taking orders?  Are you still getting 175MH/s?  Estimated pricing?  Discounts for bulk orders?

I'm not planning to sell these fully assembled -- I'd rather be designing stuff than setting up a web store, handling
end user support, etc. I might sell bare PCBs and/or write a build guide though.

I'm at about 180MH/s now in 6s150 -2 (without overclocking). Most of this progress came from ztex
(who is a Verilog and ISE-tuning God). That guy's really on a whole different level..

-rph

[amazingrando]

Any plans in the future to make larger board that have multiple FPGA sockets?

Yes absolutely.

What is the timeline for taking orders?  Are you still getting 175MH/s?  Estimated pricing?  Discounts for bulk orders?

I'm not planning to sell these fully assembled -- I'd rather be designing stuff than setting up a web store, handling
end user support, RMAs, etc. I might sell bare PCBs and/or write a build guide though.

I'm at about 180MH/s now in 6s150 -2 (without overclocking). Most of this progress came from ztex
(who is a Verilog and ISE-tuning God). I thought I was pretty good.. but that guy.. god d*mn.

-rph

I like the idea of a guide and/or selling the PCB.  It's a lot more fun (and meaningful) if you can build it yourself.

It seems like the FPGA is costing around $175.  What do you expect the all-in cost to be?

For the multiple FPGA what do you think is the limit?  Could you have 16, 32, 512??

[Dexter770221]

Nice, I've had similiar plans.  But I wanted to place goldpins (2 rows) only on two sides of the board. On one side power and programming pins, on opossite side few I/O's. Then angle goldpins should allow to place boards like cards into motherboard, or straight one just like in your idea (2 possibilities of doughterboard).
How many of those goldpins are for power? Didn't you notice significant voltage drop?

[sadpandatech]

I like the idea of a guide and/or selling the PCB.  It's a lot more fun (and meaningful) if you can build it yourself.

It seems like the FPGA is costing around $175.  What do you expect the all-in cost to be?

For the multiple FPGA what do you think is the limit?  Could you have 16, 32, 512??

  This.

  And, any plans to daisy chain the usb so there is less to plug into the carrier? Even 10 to 1 cabling would save a lot of headache.

  Looks friggin aweome and mad props to Ztex on the Verilog work.

[Transisto]

The next step is a carrier that accepts a crapload of these FPGA modules.

+1 for crapload, do want !

[Valalvax]

Btw... what is the full cost for a running miner right now?

Still just 175?

[rph]

Still just 175?

It's more than $175 per FPGA in qty 1. But significantly lower if you're building a 10GH+ cluster.

-rph

[rph]

Nice, I've had similiar plans.  But I wanted to place goldpins (2 rows) only on two sides of the board. On one side power and programming pins, on opossite side few I/O's. Then angle goldpins should allow to place boards like cards into motherboard, or straight one just like in your idea (2 possibilities of doughterboard).
How many of those goldpins are for power? Didn't you notice significant voltage drop?

Yeah, I probably should have made them usable with just 2 of the 4 headers installed.
It would have made the PCB routing much harder though.

There are 6 VCC and 6 GND pins for the core supply which is more than enough for 10A+.

-rph

[gmaxwell]

It occurred to me that with this daughter-board design you could bolt a large number of the daughterboards directly to a large monolithic heatsink then attach the host board, allowing the gold fingers to eat any of the placement error.    That should eliminate the interface problems that often come up with using a single heatsink with multiple chips. (Perhaps you were planning this already?)

[rph]

It occurred to me that with this daughter-board design you could bolt a large number of the daughterboards directly to a large monolithic heatsink then attach the host board, allowing the gold fingers to eat any of the placement error.    That should eliminate the interface problems that often come up with using a single heatsink with multiple chips. (Perhaps you were planning this already?)

I looked into using a big CU or AL plate as a heatsink. It works OK for a prototype/one-off if you have
one lying around, but to machine a custom one for low vol production is actually pretty expensive. My plan right
now is to use those individual heatsinks in the pics above, with a 120mm fan per 4 FPGAs.

I definitely don't want 50 $8 northbridge heatsink/fans in a 50 FPGA cluster. Not cost effective,
they'd increase the PCB size, produce a lot of noise, and have a poor MTBF and CFM.

-rph

[aTg]

It would be interesting to publish some schemes to duplicate the same type of circuit on breadboard or soldering, surely we could come up with new ideas like a fpga interconnect modules or try extreme overclocking with liquid cooling.

I personally am much more interested in developing the technology rather than wasting time manufacturing and distributing something.

[finway]

Why is your design so cheap?  I think other ones are about 3$/Mh above. 

[sadpandatech]

Why is your design so cheap?  I think other ones are about 3$/Mh above. 

  The other designs have a lot of costs tied into the pcb licensing, more complex pcb designs and daughter boards, etc.

  His cost of $175 is what it would cost you to buy the chip yourself and build it, plus a few buks for the pcb and components.

[finway]

  The other designs have a lot of costs tied into the pcb licensing, more complex pcb designs and daughter boards, etc.

  His cost of $175 is what it would cost you to buy the chip yourself and build it, plus a few buks for the pcb and components.

thanks.

[aTg]

Why not release the circuit diagram? so could be developed between all, not everyone is interested in buying, there are those who want to work in hardware.

PD:we would like to see photos of the prototype that had first.