Official Open Source FPGA Bitcoin Miner (Last Update: April 14th, 2013) - page 48.

kokjo

legendary

Activity: 1050

Merit: 1000

You are WRONG!

Quote from: fpgaminer on May 20, 2011, 10:14:52 PM

Quote

If you just want to fool around, start with this:
http://www.xilinx.com/products/boards-and-kits/AES-S6MB-LX9.htm

It's just $89.

Yeah you can get some cheap FPGA boards no problem. Terasic sells the DE0-Nano board for $79USD, or $59 if you're a student. That's for a 22K LUT device, which is pretty darn good for the price!

Quote

I'd respect you if you got 1 MH/s out of it.

You can get 1 MH/s out of it no problem.

can any FPGA work? how small can it be? will this be enogth: http://cgi.ebay.com/EP2C5T144-Altera-CycloneII-FPGA-mini-Development-Board-/270748203437?pt=BI_Electrical_Equipment_Tools&hash=item3f09d9b9ad#ht_1520wt_907
and how high can you clock the boards? without they are catching fire.

it would be nice to have a 1-2Mhash/s board, just for fun.

fpgaminer

hero member

Activity: 560

Merit: 517

Quote

If you just want to fool around, start with this:
http://www.xilinx.com/products/boards-and-kits/AES-S6MB-LX9.htm

It's just $89.

Yeah you can get some cheap FPGA boards no problem. Terasic sells the DE0-Nano board for $79USD, or $59 if you're a student. That's for a 22K LUT device, which is pretty darn good for the price!

Quote

I'd respect you if you got 1 MH/s out of it.

You can get 1 MH/s out of it no problem.

allinvain

legendary

Activity: 3080

Merit: 1080

This breaks my heart, open source FPGA mining code, yet the hardware is prohibitively expensive for the performance that it gives. For $600 I can buy nearly 4 5870s!

greenlander

newbie

Activity: 28

Merit: 0

Quote from: kokjo on May 20, 2011, 06:54:18 PM

well that would be very nice, so all us with no big money could try too Cheesy

If you just want to fool around, start with this:
http://www.xilinx.com/products/boards-and-kits/AES-S6MB-LX9.htm

It's just $89.

You wouldn't be able to unroll the loop at all, though. (fpgaminer fully unrolls the loop so it's heavily pipelined.) You'd probably have re-use the same hardware for every step of the algorithm, so you'd get about 1/128th of the performance. If you were clever you might be able to do more than that and have around 4 pipeline stages or so, then you'd get 1/32th of the speed.

I'd respect you if you got 1 MH/s out of it.

Still, it could be a fun project for $89. Just don't expect to do any serious mining.

It appears to come with a crippled version of their tool chain. However, you can download the full Xilinx tool chain at (ahem) substantially below retail value over bittorrent.

kokjo

legendary

Activity: 1050

Merit: 1000

You are WRONG!

Quote from: fpgaminer on May 20, 2011, 06:42:41 PM

Quote

A Spartan 3E 250K has 5508 logic cells. If I understand correctly, this is not enough for a design that needs 90K LUTs, though I'm still learning about all of this.

I'll try to put my serial design online as well. That fits into 3K or 4K I think, although it's obviously much slower (64 cycles per hash). It's a nice toy for people to play with if they have smaller boards.

well that would be very nice, so all us with no big money could try too Cheesy

fpgaminer

hero member

Activity: 560

Merit: 517

Quote

A Spartan 3E 250K has 5508 logic cells. If I understand correctly, this is not enough for a design that needs 90K LUTs, though I'm still learning about all of this.

I'll try to put my serial design online as well. That fits into 3K or 4K I think, although it's obviously much slower (64 cycles per hash). It's a nice toy for people to play with if they have smaller boards.

teknohog

sr. member

Activity: 520

Merit: 253

555

Quote from: ataranlen on May 20, 2011, 05:36:44 PM

I'm probably going to pick up a Spartan 3E 3 E 250K, so i'll jump on this FPGA bandwagon as well!

A Spartan 3E 250K has 5508 logic cells. If I understand correctly, this is not enough for a design that needs 90K LUTs, though I'm still learning about all of this.

ataranlen

hero member

Activity: 846

Merit: 1000

The One and Only

I'm probably going to pick up a Spartan 3E 3 E 250K, so i'll jump on this FPGA bandwagon as well! You just made getting started alot easier!

fpgaminer

hero member

Activity: 560

Merit: 517

First off, many, many thanks to the person who made a donation while I was asleep. I'm assuming it was because of this project? Either way, you have my gratitude! It makes me feel that my efforts on this project are appreciated.

Quote

just a optimision, from a noob:

kokjo, thank you for your suggestion! I appreciate all feedback on this project.

Quote

is it possible in sha256_transform.v to remove the generate loop, and replace it with some sort of pipelining instead

That is indeed what the Verilog compiler does with a generate loop. It unfolds the loop, replicating the logic. So that sha256_transform module actually contains 64 instances of sha256_digester.

Quote

btw is it transmiting all the hash, if it is, wouldnt it be better to transmit a signal if the H was 0?

Hello ktion23. Thank you for your feedback and interest! It only transmits hashes where H is 0. In fpgaminer_top, this is the code that accomplishes that:

Code:

is_golden_ticket <= hash2[255:224] == 32'h00000000;
old_golden_ticket <= is_golden_ticket;
if(is_golden_ticket && !old_golden_ticket)
begin
golden_nonce <= nonce;
end

Speaking of which, I need to clean that up a little bit. old_golden_ticket isn't necessary at all. I'll go do that now ...

EDIT: On a more personal note, this past week has been pretty rough. Some bad mojo going around the company I work for, which is always stressful. But seeing this interest in the project, feedback, and that donation this morning have made be feel a lot better. So ... simply, thank you!

ktion23

newbie

Activity: 1

Merit: 0

Quote

just a optimision, from a noob:
is it possible in sha256_transform.v to remove the generate loop, and replace it with some sort of pipelining instead, it will from what i can understand make the code smaller, but have a greater delay. and with smaller code you could have more of them on a single board, and a higher clockspeed.

am i talking gibberish, or do i have something right?
its was all from what i could learn from the internet in about 5 hours. Smiley

i just think it would be an improvement

is it posible to "restart" the calculus,

btw is it transmiting all the hash, if it is, wouldnt it be better to transmit a signal if the H was 0?

gusti

legendary

Activity: 1099

Merit: 1000

Quote from: greenlander on May 20, 2011, 11:24:45 AM

Quote from: gusti on May 20, 2011, 07:44:05 AM

Sorry, I do not see your point that power over capital investment cost changes with difficult. Can you please ellaborate or give a practical example ?

Today, you buy $2000 worth of hardware. Your hash rate is 1500 MH/s. You draw 1200 watts. A bitcoin is worth $7. Electricity is $0.10/kWh. You gross $1228 a month and your electricity cost is $86 a month. Your electricity cost is just noise.

Imagine that a year from now, the difficulty is tenfold.

At this point a FPGA rig is going to have a fraction of your power consumption. At that point:
Your hash rate is 1500 MH/s. You draw 1200 watts. You gross $123 a month and your electricity cost is $86 a month. Your net profit is $37/month. Your electricity cost consumes most of your gross.

The guy with the FPGA array has the same hash rate of 1500 MH/s. He draws 300 watts. He grosses $123 a month and his electricity cost is $21 a month. His net profit is $102 a month.

Will difficulty rise tenfold? Will bitcoins continue to be worth $7? Will 1500MH/s of hashing power only draw 300 watts in FPGAs? Who knows? I'm just pointing out the scenario where the FPGA guy can have a huge advantage.

In your example, you are overriding the initial FPGA costs. At this moment, for a 1500 Mhps FPGA array, you need 18.7 x 80 Mhps FPGA, or $ 11100.- Same 1500 Mhps is achieved with $2000 worth GPU's. So, assuming a profit gain for FPGA's of 102-37 = 65 a month, you still need almost 12 years to recover initial FPGA investment.

greenlander

newbie

Activity: 28

Merit: 0

Quote from: gusti on May 20, 2011, 07:44:05 AM

Sorry, I do not see your point that power over capital investment cost changes with difficult. Can you please ellaborate or give a practical example ?

Today, you buy $2000 worth of hardware. Your hash rate is 1500 MH/s. You draw 1200 watts. A bitcoin is worth $7. Electricity is $0.10/kWh. You gross $1228 a month and your electricity cost is $86 a month. Your electricity cost is just noise.

Imagine that a year from now, the difficulty is tenfold.

At this point a FPGA rig is going to have a fraction of your power consumption. At that point:
Your hash rate is 1500 MH/s. You draw 1200 watts. You gross $123 a month and your electricity cost is $86 a month. Your net profit is $37/month. Your electricity cost consumes most of your gross.

The guy with the FPGA array has the same hash rate of 1500 MH/s. He draws 300 watts. He grosses $123 a month and his electricity cost is $21 a month. His net profit is $102 a month.

Will difficulty rise tenfold? Will bitcoins continue to be worth $7? Will 1500MH/s of hashing power only draw 300 watts in FPGAs? Who knows? I'm just pointing out the scenario where the FPGA guy can have a huge advantage.

Quote from: Basiley on May 20, 2011, 03:52:03 AM

but at that "reach point" you need FPGA-utilising solution with multiple FPGA's per board or use them into mini-boards[ISA bus ? Tongue

], stacked into main one, to minimise communication/power/connectivity impact on solution cost.
and also, board must be ready to put into industry-adopted-sizes racks[both IT and manufacturing/science racks meant]. if you plan to sell number of such boards.

Well, yeah. I don't think that anyone is proposing that FPGA reference boards are the way to go. The way to go is to do your own board design and load it up with bucketfuls of FPGAs.

kokjo

legendary

Activity: 1050

Merit: 1000

You are WRONG!

just a optimision, from a noob:
is it possible in sha256_transform.v to remove the generate loop, and replace it with some sort of pipelining instead, it will from what i can understand make the code smaller, but have a greater delay. and with smaller code you could have more of them on a single board, and a higher clockspeed.

am i talking gibberish, or do i have something right?
its was all from what i could learn from the internet in about 5 hours.

i just think it would be an improvement

gusti

legendary

Activity: 1099

Merit: 1000

Quote from: greenlander on May 20, 2011, 03:41:50 AM

Quote from: gusti on May 19, 2011, 11:07:13 PM

Nice, and many thanks for the release.
Now, some rough calculations.

At 80 MHps, I will need at least 3 of these to achieve a single 5830 hashrate.
That is $595.-x 3 = $1785.- at full price, vs. $190.- for the 5830.

Giving the 5830 is consuming $11.- a month in electricity, and assuming this board will consume zero electricity, it will take more than 145 months, or 12 years to recover the investment, always comparing to a 5830.

Is it really cost feasible ?

Your math assumes the difficult will be constant.

Eventually, the difficulty will reach a point where the power cost is a significant portion of the mining cost. Right now, power is about 10%-20% of the cost when doing GPU bitmining.

Let's assume that one year from now, power is 80% of the mining cost. At that point, FPGA mining would have a significant advantage over GPU mining even if the hash rates are same.

Sorry, I do not see your point that power over capital investment cost changes with difficult.
Can you please ellaborate or give a practical example ?

eturnerx

member

Activity: 84

Merit: 10

That's fpgaminer for opensourcing the code. Imma start hacking away!

Jaime Frontero

full member

Activity: 126

Merit: 100

Quote from: greenlander on May 20, 2011, 03:41:50 AM

Quote from: gusti on May 19, 2011, 11:07:13 PM

Nice, and many thanks for the release.
Now, some rough calculations.

At 80 MHps, I will need at least 3 of these to achieve a single 5830 hashrate.
That is $595.-x 3 = $1785.- at full price, vs. $190.- for the 5830.

Giving the 5830 is consuming $11.- a month in electricity, and assuming this board will consume zero electricity, it will take more than 145 months, or 12 years to recover the investment, always comparing to a 5830.

Is it really cost feasible ?

Your math assumes the difficult will be constant.

Eventually, the difficulty will reach a point where the power cost is a significant portion of the mining cost. Right now, power is about 10%-20% of the cost when doing GPU bitmining.

Let's assume that one year from now, power is 80% of the mining cost. At that point, FPGA mining would have a significant advantage over GPU mining even if the hash rates are same.

my power cost right now is about 4-5% of what i mine in Bitcoin, at $7/1BTC. 1 - 1 1/2 days of mining pays my monthly electric bill (that 80% of it which is incurred by mining). my rigs are paid for. it's not as grim - and won't be - as you make it out, IMHO.

Basiley

newbie

Activity: 42

Merit: 0

Quote from: greenlander on May 20, 2011, 03:41:50 AM

Quote from: gusti on May 19, 2011, 11:07:13 PM

Nice, and many thanks for the release.
Now, some rough calculations.

At 80 MHps, I will need at least 3 of these to achieve a single 5830 hashrate.
That is $595.-x 3 = $1785.- at full price, vs. $190.- for the 5830.

Giving the 5830 is consuming $11.- a month in electricity, and assuming this board will consume zero electricity, it will take more than 145 months, or 12 years to recover the investment, always comparing to a 5830.

Is it really cost feasible ?

Your math assumes the difficult will be constant.

Eventually, the difficulty will reach a point where the power cost is a significant portion of the mining cost. Right now, power is about 10%-20% of the cost when doing GPU bitmining.

Let's assume that one year from now, power is 80% of the mining cost. At that point, FPGA mining would have a significant advantage over GPU mining even if the hash rates are same.

but at that "reach point" you need FPGA-utilising solution with multiple FPGA's per board or use them into mini-boards[ISA bus ? Tongue

], stacked into main one, to minimise communication/power/connectivity impact on solution cost.
and also, board must be ready to put into industry-adopted-sizes racks[both IT and manufacturing/science racks meant]. if you plan to sell number of such boards.
IMO.

greenlander

newbie

Activity: 28

Merit: 0

Quote from: gusti on May 19, 2011, 11:07:13 PM

Nice, and many thanks for the release.
Now, some rough calculations.

At 80 MHps, I will need at least 3 of these to achieve a single 5830 hashrate.
That is $595.-x 3 = $1785.- at full price, vs. $190.- for the 5830.

Giving the 5830 is consuming $11.- a month in electricity, and assuming this board will consume zero electricity, it will take more than 145 months, or 12 years to recover the investment, always comparing to a 5830.

Is it really cost feasible ?

Your math assumes the difficult will be constant.

Eventually, the difficulty will reach a point where the power cost is a significant portion of the mining cost. Right now, power is about 10%-20% of the cost when doing GPU bitmining.

Let's assume that one year from now, power is 80% of the mining cost. At that point, FPGA mining would have a significant advantage over GPU mining even if the hash rates are same.

fpgaminer

hero member

Activity: 560

Merit: 517

Quote

Do you have any data from Altera's tool chain on what % of the hardware units you used? (perhaps it could fit into a smaller device?)

This version uses ~90K LUTs. I have a more optimized version in the works (<80K LUTs), but it's still being tested and tweaked.

Jaime Frontero

full member

Activity: 126

Merit: 100

Quote from: at0m on May 20, 2011, 12:51:45 AM

Looks promising. A pretty decent hash speed to start with. Can't wait to see where the community takes this.

indeed.

let us keep in mind that when the Bitcoin client was first released, the very idea of even 20 Mhash/second was like some far-off holy grail.

80 Mh/s? that's pretty damn stout for a first whack...

Topic: Official Open Source FPGA Bitcoin Miner (Last Update: April 14th, 2013) - page 48. (Read 432921 times)