FPGA mining

[Grinder]

6950 level cards will drop to around $100.  

That's fairly unlikely. 5850 came out 1.5 years ago. Except for the somewhat peculiar Sapphire 5850 Xtreme, the price never fell much below $200.

[1715158497]

1715158497

[Littleshop]

6950 level cards will drop to around $100.  

That's fairly unlikely. 5850 came out 1.5 years ago. Except for the somewhat peculiar Sapphire 5850 Xtreme, the price never fell much below $200.

Maybe this should be put into the betting pool.

Something like.. major online shop like newegg or tigerdirect to have a card readily available (not just a fluke sale) that can hash at 240mh/s for $100 by 10/1/2012

When I say 6950 level I mean a card that can hash at a similar level, I do not mean the exact card.  They may come out with a new line lets say the 7000 series with a model that fits in at that hashing performance.  I do think in 1.5 years it could happen. 

[allinvain]

So anyone selling an FPGA miner box yet?

[fpgaminer]

So anyone selling an FPGA miner box yet?

Sure, if you can afford the steep cost of FPGAs.

I haven't much thought about an actual product, since I doubt anyone would buy it at the mhash/$ rate FPGAs get.

[fpgaminer]

Huzzah! I got my LaunchPad microcontroller kits today. $4 USD each, shipped.  Too bad TI's development tools have terrible licenses. Only free one is code-size limited 

I'll be putting them to use as controllers for the FPGAs; communicating with a pool over ethernet and providing a remote interface. This will make the FPGAs completely independent, and far more stable than a GPU miner 

Only thing I'm missing are ethernet controllers. I'm looking at WIZnet W5100 breakout, but will have to wait until I've mined enough bitcoins to buy them 

P.S. Yes, I know the Altera C120 dev kit has an on-board ethernet controller. And no, I don't want to lose all my hair fighting SOPC system builder to get it working.

[allinvain]

So anyone selling an FPGA miner box yet?

Sure, if you can afford the steep cost of FPGAs.

I haven't much thought about an actual product, since I doubt anyone would buy it at the mhash/$ rate FPGAs get.

Depends. How steep are these costs? Toss around some $ figures.

[fpgaminer]

How steep are these costs? Toss around some $ figures.

Using what I've developed to-date, these would be the costs (USD):

DIY Kit - 80MH/s, computer required

* Altera DE2-115 Development Board - $600

Total: $600



Fairly soon I hope to have the LaunchPad micros up and running. So that'd be:

Future DIY Kit - 80MH/s, runs independently (no computer)

* Altera DE2-115 Development Board - $600
* LaunchPad eval kit - $4
* Ethernet module - $25

Total: $629


Supposing a custom board was made:

Possible Custom FPGA Board - 80MH/s, runs independently (no computer)

* Altera Cyclone 4 C115 chip - $400
* Custom PCB & parts - $20 - $50?
* Micro - $4?
* Ethernet - $15?

Total: $469


Anything with a question mark is an educated guess. And of course a custom board has design costs associated with it, which would have to be folded into any quantity of units sold.

[allinvain]

Definitely not cheap. It is as you say, the cost per Mhash rate is definitely not worth it.

I'm also wondering if 80 Mhash is the max performance you can squeeze out of it?

[Sukrim]

The big question would be power draw + coolability then...

If they are easy to cool, it might be much easier to stack them/have more of them at once.
If they draw much less power per MH/s, in the longer run it might pay off against a graphics card (longer run = maybe years!)

[fpgaminer]

If they draw much less power per MH/s

The 80Mhash/s design draws less than 5 Watts. With a micro and Ethernet that might be a tiny bit more, but not much. The power regulator will also take it's share. So, I dunno, let's call it 10 Watts. They will basically always be profitable. It just might take a year to pay them off first.

If they are easy to cool, it might be much easier to stack them/have more of them at once.

The 80Mhash/s design requires cooling from a small fan. I use a small 5 Watt fan and it does the job. You could probably get smaller, or even just use a passive heatsink.

Down clocked to 50Mhash/s and it doesn't even need cooling at all. You could easily stack them, with one micro controller driving several FPGAs.

I'm also wondering if 80 Mhash is the max performance you can squeeze out of it?

80 Mhash/s is the best I've gotten so far. My next goal is 160 Mhash/s inside a slightly larger FPGA. I'm by no means a guru here. Someone with more experience than I could easily redo the adder chains and possibly redo the register spacing to pull maybe 120 Mhash/s out of the chips I quoted above.

[caston]

Have you thought about some kind of hybrid GPU / FPGA solution?

e.g. http://www.siliconwolves.net/frames/projects.html

[newMeat1]

This sounds cool. I'm an EE and I took an FPGA class in the past, and I've printed custom circuit boards. For a small batch of circuit boards, the cost is about $80. That would be enough for at least 3 circuits, maybe more (depending on how big they are).

If somebody helps me with FPGA choice and programming, I could *probably make custom boards, cheap. I think this is a good business opportunity.

If anybody wants to work on this over the summer, PM me. I have about 2 months of free time... just enough to get something done, I think.