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

[aTg]

Honestly, this is the only serious and well thought out project that exists now to create a cluster bitcoin FPGA.
I hope you continue to finish up, we are eager to reproduce your configuration somehow.

[rph]

we are eager to reproduce your configuration somehow.

heh, the more the merrier!

-rph

[worldinacoin]

What is that program you are running?  Looks neat and cool!

FPGA mining has come a long way since September.
Arsbitcoin needs to fix their service  but with multi-pool support it's not a big issue.
Thanks again to TheSeven for his awesome miner SW.

-rph

[rph]

What is that program you are running?  Looks neat and cool!

It's called mpbm and it's GPL'd. It's pretty sweet!

-rph

[ngzhang]

What is that program you are running?  Looks neat and cool!

It's called mpbm and it's GPL'd. It's pretty sweet!

-rph

please give me a link... i cannt find it...

[sadpandatech]

 Heya RPH, any good news on the multi carrier design?

  Cheers

[aTg]

We have to all like a child waiting for Santa Claus to see your working quad-core design.
Tip: You could put a power connector of this type to facilitate the use of ATX.

And also use a simple pin to usb conection, so you can use a regular cable and connect the circuit to the PC motherboard.

Greetings, new hope

[rph]

Thanks guys. There's some crazy stuff in the works! 

-rph

[plastic.elastic]

We have to all like a child waiting for Santa Claus to see your working quad-core design.
Tip: You could put a power connector of this type to facilitate the use of ATX.

And also use a simple pin to usb conection, so you can use a regular cable and connect the circuit to the PC motherboard.

Greetings, new hope

Negative, what a silly idea.

You limit yourself with a desktop components that use ATX PSU and USB pin header?

What about mobile computers?

[aTg]

This will be all right? trying to reduce production costs of the pcb, then each can adapt it to specific cases.

[Albert]

This is very nice.  

I live off peanuts though (Like some/alot of the people fascinated with Bitcoins), so are you sure this will go cheap? (I realize I may contradict myself a bit in another FPGA thread, but I am assuredly willing to chip in the money to see this through)

EDIT: Hmm, could you possibly sell the PCB specs of the FPGA and carrier, under a non-redistributable license? Or is home-fabrication of these still a pipedream? With traces being as thin as they are.

[heavyb]

have you decided to sell these yet?

If you're interested in a 10GH/s+ cluster - that could start to make sense - PM me and we can discuss.
I don't want FPGAs to become too widely available.. I want to compete with 45nm GPUs forever  

-rph

haha nah, I was thinking more like one board. I would like to learn to reproduce them as well. I am a Web Design and Programming teacher, and learning this interests me.

[hardpick]

Skillet-reflowed and ultra cheap. No need to pay $500+ for an eval board or $50+ per chip to an assembly house;
high performance FPGAs can be soldered at home in 5 minutes with some flux and a $20 skillet.

It's a xc6slx25 on there now to validate the PCB + asm process; building xc6slx150 soon.







-rph

sorry for the dumb question
where do I buy the two boards?

[DeathAndTaxes]

Skillet-reflowed and ultra cheap. No need to pay $500+ for an eval board or $50+ per chip to an assembly house;
high performance FPGAs can be soldered at home in 5 minutes with some flux and a $20 skillet.

It's a xc6slx25 on there now to validate the PCB + asm process; building xc6slx150 soon.







-rph

sorry for the dumb question
where do I buy the two boards?

You don't.  They aren't for sale.    I wish rph would offer it for sale though.  The daughter board setup would be useful for making much larger boards (something no other FPGA developer seems interested).  For example he could design a 6 socket board powered by a single PCIe connector and then the daughter boards seperately.  So say $100 for the base board and then $175 per FGPA socket.  You could expand your miner from 175MH to 1GH as your budget allows.

Alas rph has no interest in retail product it seems.

[DeepBit]

For example he could design a 6 socket board powered by a single PCIe connector and then the daughter boards seperately.

I don't think anyone is going to offer PCI or PCIe solutions at least in 2012 because
1) It will make the board much more complicated.
2) It will require more complex drivers.
3) It will require a computer with enough expansion slots. Mostly dedicated miners are going to use FPGA/ASIC hardware, not your average gamers/overclockers, so it's not like "just add a board", they will need many of them with as small overhead expenses as possible. Also building big farms with only a few boards per case is not space-efficient.

Why place a board inside if one can just connect an arbitrary number of external boards ?
But actually I don't like USB and can expect Ethernet-connectible solutions, may be even with autonomous mode, not requiring a computer at all.

[it_guru]

For example he could design a 6 socket board powered by a single PCIe connector and then the daughter boards seperately.

I don't think anyone is going to offer PCI or PCIe solutions at least in 2012 because
1) It will make the board much more complicated.
2) It will require more complex drivers.
3) It will require a computer with enough expansion slots. Mostly dedicated miners are going to use FPGA/ASIC hardware, not your average gamers/overclockers, so it's not like "just add a board", they will need many of them with as small overhead expenses as possible. Also building big farms with only a few boards per case is not space-efficient.

Why place a board inside if one can just connect an arbitrary number of external boards ?
But actually I don't like USB and can expect Ethernet-connectible solutions, may be even with autonomous mode, not requiring a computer at all.

So what about using a http://beagleboard.org/bone as your host/computer.  It has Ethernet and USB it might be a little slow for a python or tcl "controll" software but in c it should be fast enough.

-IT

[DeepBit]

So what about using a http://beagleboard.org/bone as your host/computer.  It has Ethernet and USB it might be a little slow for a python or tcl "controll" software but in c it should be fast enough.

If you need just one then it's fine, but for larger quantities onboard ethernet controller would be cheaper.

[DeathAndTaxes]

For example he could design a 6 socket board powered by a single PCIe connector and then the daughter boards seperately.

I don't think anyone is going to offer PCI or PCIe solutions at least in 2012 because
1) It will make the board much more complicated.
2) It will require more complex drivers.
3) It will require a computer with enough expansion slots. Mostly dedicated miners are going to use FPGA/ASIC hardware, not your average gamers/overclockers, so it's not like "just add a board", they will need many of them with as small overhead expenses as possible. Also building big farms with only a few boards per case is not space-efficient.

Why place a board inside if one can just connect an arbitrary number of external boards ?
But actually I don't like USB and can expect Ethernet-connectible solutions, may be even with autonomous mode, not requiring a computer at all.

I think you misunderstood.  I was speaking of the ATX PCIE power connector.  It can supply up to 75W which should be sufficient for 8 FPGA.  Given most large clusters will be powered by a high efficiency ATX power supply having a larger board w/ support for multiple chips powered by a single plug makes sense right?  Nobody is going to want 50+ dinky low efficiency wall wart power supplies (at $10 ea = $500) to power a large array.

Take ztek board for example.  He provide instructions to mod an ATX PCIe power connector to 6 single DC plug connectors and then connect them to 6 individual boards.   Now imagine instead of doing that you have:

A) a "motherbaord" w/ 6 FPGA sockets, an PCIe 6pin power connector, an onboard motorolla ARM CPU (think rasberry Pi) and ethernet connectivity.

B) individual daughtercards for 1 FPGA each.

A single 80-Plus Gold PSU could power 4 to 8 of these boards.  Each board could be loaded w/ 1 to 6 (maybe FPGA daughter cards.  Giving you a high hashing density.

Alternatively instead of having the CPU on the board you could make it a separate board (Rasberry Pi would work and is only $35) and connect to multiple FPGA boards via USB or serial interface.

[DeepBit]

I think you misunderstood.  I was speaking of the ATX PCIE power connector.  It can supply up to 75W which should be sufficient for 8 FPGA.  Given most large clusters will be powered by a high efficiency ATX power supply having a larger board w/ support for multiple chips powered by a single plug makes sense right?  Nobody is going to want 50+ dinky low efficiency wall wart power supplies (at $10 ea = $500) to power a large array.

Oh, you were talking about power, it makes sense then.
Actually the more important is the low-voltage supply. ATX PSU can only provide most of it's power as 12 V and FPGAs need something around 1.2V, so additional regulators are required. But using one 1.2 regulator for many boards will require expensive high-current connectors, so multiple chips per board would be effective solution (of course if you won't need to replace bad/faulty chips :)

[DeathAndTaxes]

Oh, you were talking about power, it makes sense then.
Actually the more important is the low-voltage supply. ATX PSU can only provide most of it's power as 12 V and FPGAs need something around 1.2V, so additional regulators are required. But using one 1.2 regulator for many boards will require expensive high-current connectors, so multiple chips per board would be effective solution (of course if you won't need to replace bad/faulty chips

Yeah that was my thinking.  Most (all) of the boards I have seen take 12V DC and then use internal regulator to drop that down to voltage used by FPGA.  Using one voltage regulator per FPGA "socket" is like a good design decision.  It means a failure would "only" destroy one FPGA.  Given the low cost of 10W 12V to ~1.2V power supplies and the high cost of FPGA "compartmentalizing" the board is good insurance.

Even at $1 per MH.  A 1GH board is ~$1000.  I would feel more comforable driving $1000 in electronics off a high quality ATX power supply than some wallwart made in China at the lowest possible cost.  If we are taking about an array of say 6x 1GH boards (300W total) I wouldn't want to use anything other than an ATX powersupply or better.

I think if someone could make a 6 socket board which is less than 15" by 7" you could mount them in a 19" 4U industrial rack w/ 3" spacing between boards.  They make "industrial" chassis which are nothing but expansion slots (no motherboard).  They have a 5x 90mm set of fans in the midplane which provides uniform cooling.  Airflow is front to back and the nicer chassis have redundant fans (2x5 90mm).  This wouldn't be a traditional "PC" but would provide a safe method for racking up cards.  Given the cost of a large array ($6000 for GH) I would want some physical and electrical protection rather than just cards laying on a desk.  Hell w/ the low power consumption you could put it in a datacenter and insure it against loss as any other industrial asset.  ~6GH in a 4U rack. 60GH per data center rack and it would only draw 12 amps on a 240V circuit.