I wonder that myself.  I wonder how large it will need to be not because the FPGA take up a lot of space but due to the fact that 2.5KW is a lot of energy and the smaller the pkg the harder it is to cool.  

A visual excercise:
A 4x5970 rig is ~1KW.  Open framed it requires good amount of airflow to cool.   In a closed case, forget about it.  Now 1.25KW would be 20% more thermal load so image a hypothetical 5x5970 rig (pretend no 8 GPU limit). The "full rig box" (if it ends up getting built) would be double that.   To get an idea of the thermal energy vs volume visualize a 4U server chassis w/ 5x5970s inside now stack two of them on top of each other to simulate 2.5KW of power draw.  That is a pretty big box (8U = 14" by 19" by 24").  Could you cool that with air?
 How noisy would it be?  We already have a pretty large "rigbox".  Going larger would make it somewhat easier to cool.

Before I get flamed:
I am not saying the rigbox isn't more efficient.  It is.  The same sized box with the same amount of electricity/heat would produce 50 GH/s vs ~7 GH/s (10x 5970s). Still the laws of thermodynamics don't care what (or how much) the 2.5KW produced.  50GH/s on 2.5KW or 7GH/s on 2.5KW is still 2.5KW either way.  

That is an interesting challenge.  In mining the difficulty in cooling 1KW+ heat loads is what has lead to open frame rigs, extenders, extra high rpm fans, etc.  Will be interesting to see how they intend to cool them and how noisy they will be.  Not saying high potential noise is a problem (it isn't as most will be in datacenters or semi-industrial locations).  It just may shock some people who think of FPGA as silent and cool but that is just because nobody has scaled FPGA up to tens of GH/s in a compact space yet.

So BFL come on at least give us some estimated dimensions and db specs.

As someone who was an outspoken (and loud) critic of BFL I am convinced they aren't a "scam".  They masively overestimated their specs and timelines.  They also seem ill equipped to handle the non-tech portions of the demand they are facing.

IMHO the only real risks dealing w/ BFL are:
a) time risk.  800 MH/s miner earns ~15 BTC per month.  4 month delay = 60 BTC in lost revenue.
b) warranty risk.  If new orders take 12-16 weeks then how long will RMA take.
c) solvency risk. If company had a huge financial setback (say 1000 singles died under warranty) could they absorb that and continue to operate

basically the risks are the same as dealing with any small business where demand way outstrips productive capacity.  It would be smart for BFL to hire an business student (for next to nothing) to handle non-tech tasks like emails, order processing, releasing company statements, etc.

Psst BFL there is 10% unemployment right now (20%+ for college age workers), labor is cheap.

What is the difference between revision 2 and 3? Didn't see that this was answered elsewhere.

At the risk of repeating myself, my point wasn't about the difficulty *achieving* the 51% attack, but more about the power that would come once it's achieved.  However, in answer to your reply, I think that you are being a bit idealistic.  What you say is perfecty possible, but perhaps perhaps not so practical.  It requires either:

• that lots of small time *individual* miners agree to firewall abusive IPs (and here I'm assuming that there are only a few abusive IPs, not millions such as BigGovt might muster)  OR
  
• that the large mining pools are not corruptible (as in, do as BigGovt says or face the chop).  And here bear in mind that the future of bitcoin is lots of non-mining thin clients, and very few large multi-GPU,FPGA,ASIC mining arrays.  Even now the major pools supply 76% of the blocks (http://blockchain.info/pools).

And if you want proof, how come nobody has blocked the IP in this story yet?  - it's creating empty blocks: https://www.privateinternetaccess.com/blog/2012/03/bitcoin-war-the-first-real-threat-to-bitcoin/

And again, like I say, if you decide to reject blocks that leave valid transactions by the wayside, you risk losing out on approved transactions to you!  You'd probably have to be running two clients - one for the Govt-approved blockchain, and another for the blackmarket chain.  Hey, the "BlackChain".  Dudes, I should copyright that 

Oh, this thread will probably be a little cleaner from now. Regarding the BFL single - I think it only uses twice the w/MH traditional FPGAs use. Wasn't it ~80 Wh for ~800 MH? Icarus and x6500 use 40 Wh for that amount of hashing.

Are people expecting the secondary market to be flooded with GPU's once singles are being easily accessible?

@BFL-Engineer: Has this been brought to your attention and are you working on a solution?

That is the 1 million BTC question.

While reward going down doe cut future supply and should be bullish to prices how much and how quickly is hard to gauge.  Remember the existing 10 million coins will still be available for trade.  Revenue in BTC will definitely be falling 50%.  Prices may rise but they likely will rise less than the 100% necessary for an "even trade" and the rise is more likely to be felt in the long term.

A miner in the short term likely will see a significant drop in revenue.  If block goes from 50 * $5 = $250 (today) to 25 * $7 = $175 after the cut the fact that prices will rise in the future may be of little comfort for the marginal miner.

Simple way to avoid that is to not be the marginal miner.  Marginal miners get squeezed out and difficulty falls boosting revenue of those who "survived".  Don't be the marginal miner.

... before the 50 to 25 BTC block reward wipes out many miners' profitability (i.e. mine - anyone in Europe with expensive electricity and GPUs will be underwater with 25 BTC block reward)...

Fred,

Can you give us more details on the throttling? Is it heat related?

I could do that, problem is i need to charge some profit. Also i need to charge VAT 23% when selling to EU resident.
So the price would end up being something like 850$

ZTEX have a guide for adapting to a computer PSU:

http://wiki.ztex.de/doku.php?id=en:ztex_boards:ztex_fpga_boards:cluster_power_supplies

I guess it can be applied here, not sure about the power though.

OK, here's one: This was after I had sold my 2-seater Cessna 150, and while I was renting planes. Arriving at the airport, I find out that all 2-seaters, which I was intimately familiar with, were rented out or in the shop, so I decide to suck it up, pay a little bit more and rent a 4-seater (Skyhawk) instead. I had been checked out in one during the BFR a few months ago, but it wasn't what you call "fits like a glove". So I take off, fly to, whatever, Monterey or something, enjoy the coastal scenery and the toy houses below and finally get back to the home airport and land. Turns out, being not 100% familiar with a Skyhawk, I land going too fast. So what does an airplane do when you land at too fast a speed? It does what it's designed to do, it takes off again. A little. It starts to hop down the runway, doing these little bunny hops. Would be a fun thing to do (the hops don't damage the landing gear or anything), except for the insignificant little fact that while you're hopping down the runway, you cannot brake. So the shopping mall conveniently located at the end of the runway starts to get larger and larger in the windshield, as I'm still hopping down the runway at 70 knots. If you panic now, you're dead. So training and survival instinct take over, I push the throttle forward, retract the flaps and pitch the airplane into best climb angle. The shopping mall is now filling the windshield, but the airplane is light, with but one person inside, and the engine does its job and finally there is a positive climb and the airplane clears the shopping mall. Once again I enter the landing pattern, but this time I land 5 knots slower and it doesn't hop and I can brake and turn off the runway.

My point is, you haven't really lived until you have almost died once.

Hey Mr BFL man,
If we don't need the included PSU, can we get a discount?

Tentative good news.  OEM is considering the order a small run.  Not sure if it is past business or that the economy is slow.  In the past I know the have balked at runs below thousand units.  No idea on pricing yet, no need to respond saying you want one, and don't get your hopes up yet.

My contact called me back.  He asked for a wiring diagram to give to the team. Anyone have experience making wiring drawings?  
Diagram is pretty simple but my graphical skills are lacking.  If anyone has experience and wants to help let me know in thread or by PM.

Also if someone can measure the PSU barrel adapter.  Measure in mm the length of the metalic part.  I know BFL indicates 12mm but I just want to make sure.  12mm seems a very strange length for 2.5mm/5.5mm barrel plug.  Can't seem to find a part # anywhere (Molex, digikey, half dozen suppliers, etc).  Found 8mm length and 14mm length but not 12mm length.    If someone from BFL is reading this and has part # for the barrel power socket on the BFL Single board it would help finding an exact match.

Thank you for providing an important puzzle piece on how Dr. Tyrell does it.

The multiplier in the DSP48-block is not needed in SHA-256, hence what he obviously uses is the 18-bit adder
BCOUT = B + D.
He uses 30 DSP blocks, 10 per red / green / blue SHA-256 instance.
For a 32 bit adder, two 18-bit adders BCOUT=B+D are needed.
Thus, he can implement five 32-bit adders per SHA instance.

So, why not just use [slow] 32-bit ripple adders everywhere, and use a few [very fast] DSP adders in some places?

The answer is, IMHO, that he uses the fast DSP adders only where they feed into longlines.
Were he to use normal ripple adders where he feeds into longlines, the aggregate delay would limit
the design to a 5 ns clock cycle.
Using the fast DSP adders will allow this design, when properly fine-tuned, to march into 4 ns clock cycle
territory, for a total MH/s number of approximately 125 MH/s or approximately 375 MH/s per Spartan6-150.

BFL Single, watch out below.