FPGA Mining Contract

[ArsenShnurkov]

Somehow FPGA.contract is not in the list:

> Public information about companies traded on GLBSE

http://companies.herokuapp.com/

I think it's absence affects the price of shares

[1714892304]

1714892304

[1714892304]

1714892304

[1714892304]

1714892304

[notme]

Statistics site showing more fpgas now, but only a couple are running.  Status update?

[fizzisist]

Statistics site showing more fpgas now, but only a couple are running.  Status update?

The boards are here: https://bitcointalk.org/index.php?topic=40058.msg639925#msg639925

But I'm doing a lot of work setting them up and testing, so mining will be intermittent for the rest of the weekend at least. I'll post some pics of them mining away soon!

[ArsenShnurkov]

1) Why our spartan 6 is so slow ?

My test implementation without any optimization got 180Mhash/s according to the Xilinx ISE tool.

2) How our current solution compares to existing ones:
http://www.dinigroup.com/new/DNBFC_S12_PCIe.php



3) How much does Virtex 7 cost ?
"The Virtex-7 2000T the 2 million logic cell will begin initial sampling in August and November of 2011"
Is it possible to receive samples ? (somehow we already miss these intervals, but it is still possible to try to obtain samples)
Is it possible to get promotional discounts for it?

For the comparison, the price of SPARTAN 6 is

I think you should target the Xilinx spartan 6, LX150(T) :  XC6SLX150-2FGG484C
Costs about 170$ at digikey for one, but I've heard 120$ with some volume.

I am asking, because it uses 28 nm technology, which is more modern that 45 nm in Spartan 6

What will be it's approximate hash rate?

one could get higher than 1.4MH/s

it contains 13x more logical gates, so 125 * 13 ~= 1.6MH/s

4) The same question about Artix-7

Next year it seems the Artix-7 series will give us the most "bang for the buck".

"Artix-7 FPGA initial samples will ship first quarter of 2012."

5) The same question about Kintex-7
It was already shipped: http://www.eetimes.com/electronics-products/electronic-product-reviews/fpga-pld-products/4214345/Xilinx-ships-first-28nm-Kintex-7-FPGAs

[Dexter770221]

3) How much does Virtex 7 cost ?
"The Virtex-7 2000T the 2 million logic cell will begin initial sampling in August and November of 2011"
Is it possible to receive samples ?
Is it possible to get promotional discounts for it?

I am asking, because it uses 28 nm technology, which is more modern that 45 nm in Spartan 6

4) The same question about Artix-7

Next year it seems the Artix-7 series will give us the most "bang for the buck".

"Artix-7 FPGA initial samples will ship first quarter of 2012."

5) The same question about Kintex-7
It was already shipped: http://www.eetimes.com/electronics-products/electronic-product-reviews/fpga-pld-products/4214345/Xilinx-ships-first-28nm-Kintex-7-FPGAs

ad2) Price. That card have features that mining don't need and it costs twice more than needed. And propably voltage regulators are not suitable to deliver required amount of power, there's no heatsinks on chips.
ad3) Virtex7 costs 30.000$ (30 thousand !) when you buy 1000 then you get discount to 28.500. Do math yourself.
ad4) Atrixes7 will be propably the best replecement for Spartans6. Price is unknown for now but Xilinx says that it will be comparable to Spartan6.
ad5) Kintexes7 costs to much, 325T part is something around 1200$ and perfomance with good design should reach 1GH/s so lower MH/$ than S6.

[ArsenShnurkov]

Do math yourself

Virtex-7:
$30000 / 1600 MHh/s = $18.75 / MHh/s

Spartan-6:
$170 / 125 MH/s = $1.36 / MHh/s

Kintexes-7:
$1200 / 1000 MH/s = $1.2 / MHh/s

ATI 6990 = $0.89..$1.11 / MH/s

there's no heatsinks on chips

Might be there some cooling system which is not shown.
Or are the heat sinks required at all?

That card have features that mining don't need and it costs twice more than needed

Then it can be used for other purposes (and probably can be sold then new technology arrive)

[ArsenShnurkov]

Statistics site showing more fpgas now, but only a couple are running.  Status update?

There are 11 of them now. There are some additional FPGA's

fpga0    A: 37    R: 8    214 Mhash
fpga1    A: 44    R: 10    257 Mhash
fpga10    A: 46    R: 7    252 Mhash
fpga2    A: 64    R: 6    334 Mhash
fpga3    A: 46    R: 10    267 Mhash
fpga4    A: 39    R: 7    219 Mhash
fpga5    A: 44    R: 18    295 Mhash
fpga6    A: 52    R: 6    276 Mhash
fpga7    A: 38    R: 16    257 Mhash
fpga8    A: 44    R: 7    243 Mhash
fpga9    A: 48    R: 11    281 Mhash

And one question remains unanswered yet:

1) Why our spartan 6 is so slow ?

My test implementation without any optimization got 180Mhash/s according to the Xilinx ISE tool.

[fizzisist]

Yep, we're off and running! I'm currently pointing 11 boards at Eligius to make up for the lost time over the last few days while working on the code. That will go back down to the 7 you guys own soon.

Here's a screenshot of this madness: http://dl.dropbox.com/u/13472215/screenshot_11boards.png

Why our spartan 6 is so slow ?

We're running the open source code developed by fpgaminer and makomk (among others). Since then, the other guys have pushed this FPGA quite a bit further. Now that these boards are finally built and are on their way out to customers, we'll turn our focus to improving that. These FPGAs are running at 133 MHz clock rate (266 MH/s / board) right now, but there's no reason we shouldn't be able to get the same as the others have achieved with the same FPGAs.

We need to work on both the FPGA code and the mining software. The mining software (that is, the code that you run on your computer) is very alpha right now. For example, it doesn't support long-polling yet.

Like I said, this will be our priority now that the hardware is built and on it's way out the door.

[ArsenShnurkov]

1) How many additional cards can be attached to the current setup without adding another hardware (except USB-hubs)?

2) Do you have plans to redesign with Kintex-7 (28nm tech)?

Spartan-6:
$170 / 125 MH/s = $1.36 / MHh/s

Kintexes-7:
$1200 / 1000 MH/s = $1.2 / MHh/s

Kintexes-7 was already shipped: http://www.eetimes.com/electronics-products/electronic-product-reviews/fpga-pld-products/4214345/Xilinx-ships-first-28nm-Kintex-7-FPGAs

Artix-7 samples is also available from Xilinx, just contact them

3) Was the "first hand" feature discussed with GLBSE ?

[molecular]

nice monday morning news. way to go fizzisist.

one question: is that gnu screen dividing the terminal real-estate?

I use a tiling wm so I probably wouldn't use that feature... but it's good to know.

[fizzisist]

nice monday morning news. way to go fizzisist.

one question: is that gnu screen dividing the terminal real-estate?

I use a tiling wm so I probably wouldn't use that feature... but it's good to know.

Thanks, molecular! It was very exciting to see these boards come alive.

Yes that is screen, but it's only on my end. You currently need to run a separate instance of the mining software for each board, so you can manage those processes however you like. I do have plans to handle multiple boards in one software instance, like cgminer does, very soon. I'll be working on that over the next weeks, and it should be a relatively easy change.

[teek]

business as usual over here?  just wondering because the stock took a pretty good dip?

teek

[fizzisist]

business as usual over here?  just wondering because the stock took a pretty good dip?

Yeah, everything's going great actually. In fact, I've cleaned up the cable mess for our FPGA cluster and it has been running at full capacity for some time now. I posted a picture of the setup here. What you see there is 16 boards, but only 7 of them are mining for this contract. If we release more shares and raise the funds for additional boards, they will be begin earning money for the shareholders immediately (just need to redirect the workers), so this might be a good time to discuss raising a motion again.

It looks like the trading volume shot up today, so a lot of shares are moving around. Some shareholders much have decided to sell off some shares now, and they offered them at the highest bids. Seems like odd timing, because we're on track to be receiving our largest dividends yet in the next couple days. Oh well, it's a free market!

[zapeta]

Yeah, looks like some shareholders sold in the 0.66 BTC/share through ~0.40 BTC/share range.  Not a bad profit if you were able to buy in at 0.25 BTC/share and aren't interested in holding for long term.

As for more shares, if you have the boards available we may as well see if the shareholders will approve it.  I think it'd make sense to see if we can get up to 10 boards first and then re-assess to see if there is more demand.  It doesn't make sense to issue a ton of shares if they don't sell. 

One thing to consider - If the new shares have a set price and someone undercuts it, the least expensive shares will sell first and the company won't get any of the BTC.  We can get around this by doing direct transfers.  A buyer would send the BTC to an address that fizzisist provides and give him their GLBSE user ID and the shares can be transferred directly to the buyer.  This would make sure that the money is going to fizzisist for purchase of the boards and not someone who is looking to sell for a profit.  However, it can quickly become an accounting headache.  I know that JLP-BMD has had to go with direct transfers as someone undercut their IPO price with a sell order that is 0.001 BTC less than the IPO price.  Just something to consider if we issue more shares.

[Dexter770221]

Big movement on glbse. Someone filled almost all bids. Price have drop considerably.

[ArsenShnurkov]

Price have drop considerably

-53%

the least expensive shares will sell first and the company won't get any of the BTC

Why should shareholders buy at high price, when they can buy absolutely the same shares at lower price?

If market estimates shares lower than management, it is the management to blame.
The shares mechanism was invented to resolve issues between management and shareholders.

It's the work of both parties to correctly determine the price.
If shares are cheap, then management can buy out shares itself.
If company is overestimated (from management point of view) and there are good ideas,
then management should propose to issue more shares.

Right now I think that price drops because fizzisist postpone solution of questions related to shareholders.

If we release more shares and raise the funds for additional boards, they will be begin earning money for the shareholders immediately

Are you ready to start 50 additional boards? 100 boards? How many boards are necessary to mine without external pool?
How the existing shareholders will vote for this, if this will significantly decrease their influence?
What will happen if three months later another company will begin to mine with Artix-7 ?
I asked these questions earlier.

[fizzisist]

Right now I think that price drops because fizzisist postpone solution of questions related to shareholders.

If we release more shares and raise the funds for additional boards, they will be begin earning money for the shareholders immediately

Are you ready to start 50 additional boards? 100 boards? How many boards are necessary to mine without external pool?
How the existing shareholders will vote for this, if this will significantly decrease their influence?
What will happen if three months later another company will begin to mine with Artix-7 ?
I asked these questions earlier.

Sorry, you're right that I never answered some of those questions. First of all, I currently have 16 boards running, but would not be willing to switch all of those over to FPGA.contract. The extra 9 boards are constantly being switched out for a different set of boards after an hour or two of testing to be packed and shipped to customers. I was thinking that we would offer shares for one board at a time, to see how it goes. At the moment, I'm not interested in running a 50 or 100 board system as a contract. The responsibility and risk would be too much, especially under the terms of this contract (where I make no profit). I would much rather you buy the boards from us and run them yourself!

As for the next generation of FPGAs, we're certainly watching those closely. Frankly, though, I don't understand what it matters to this contract. The 7 X6500s we have are already paid for and will continue mining until they either die or the electricity cost to run them becomes more than the money earned from mining. That will most likely be a very long time from now. I strongly doubt that the release of Artix or Kintex FPGAs will drive up the difficulty so much that our profits are significantly impacted.

[ArsenShnurkov]

I'm not interested in running a 50 or 100 board system as a contract

So, we clearly see the reason for the price drop
It is the decision of management not to arrange the profitability of the business for all the parties.

You don't look at this business seriuosly, just use it as a test platform for your other business.

[fizzisist]

I'm not interested in running a 50 or 100 board system as a contract

So, we clearly see the reason for the price drop
It is the decision of management not to arrange the profitability of the business for all the parties.

I don't understand. How does the number of boards change the profit per share?

[ArsenShnurkov]

How does the number of boards change the profit per share?

1) This changes weekly profit for shareholders. If that profit is low (in absolute amounts) - it doesn't worth to own.
2) If you will run your own pool the profit per share will be higher

High price was a result of the assumption that you will expand the business