"My garage" was tongue in cheek not a serious offer however you will be competing against people who's costs are much lower.

I didn't realize the price was in Euro well that is just horrible (~$0.20 per kWh).    Like I said not bad for a datacenter but the margin on mining is going to be a lot lower.  When industrial space (think racks in a warehouse) is cheap in parts of the US and China and marginal power rates are $0.05 paying more than double just means you will be the "marginal miner".

Yeah come on.  The promise to customers is they could cancel up to the day THEIR ORDER ships.   Some people tried to cancel on the first and were told orders can't be cancelled and it is now 15 days later and they haven't received hardware.  Any way you shake it the promise to cancel ended up being utterly worthless.   They promised one thing and at the last minute changed the terms.  I mean by your logic BFL is A+ business as well.  Production began back in last November so obviously they can't allow any cancellations right?

$1500 / (14 * 24 * 30 ) = $0.15 per kWh.

Not bad for datacenter but hardly "super cheap".  You could use my "datacenter" = garage and get rates 30% lower.   Some parts of the US have power rates under $0.05 per kWh.

I would add Avalons and BFL (65nm) rigs.   They will only be operational in areas with insanely low power costs.  Say <$0.05 per kWh.



Well it isn't that ASICs would infringe upon CPU/GPU production it is simply that 22/20nm is cost prohibitve especially for a product.  Other than Intel (who uses their own fabs) the first 22nm customers are cellphone chips.  For something like a cellphone paying 3x the cost per mm2 of silicon is worth it if you can cut power consumption by 30% or more.  For an ASIC miner not so much.   Generally it takes ~3 years before a new process node has a lower cost than the prior node.  That means there won't be smaller than 28nm tech until late 2015 at the earliest.


Still even with efficiency "capped" at 28nm there is a huge spread.  An Avalon is 8.8 J/GH and a ASICMiner blade is >7.5 J/GH.  HF and Cointerra are promising ~0.8 J/GH.  Bitfury achieves roughly that at 55nm so with a die shrink to 28nm they could be even less (<0.5 J/GH?).  You are looking at a 11x (potentially 16x) spread between least efficient and most efficient hardware.  This is something we never saw in the GPU era.

Another way to look at it is when difficulty is so high that an Avalon miner with $0.05 per kWh electrical rate is underwater (electricity cost > BTC value) a HF miner with higher $0.10 per kWh electrical rates will only be spending ~ 20% of gross revenue on electricity.  At some point the least efficient miners won't make sense anywhere.

No. Didn't your mom tell you stealing is wrong?

What part of don't violate the social contract is hard to understand?

If people like the OP had their way this would be a future ad for Bitcoin.


Maybe the "new-Bitcoin" slogan can be "Bitcoin because its worse than a bank".

I second jjimm_64 questions.  Some detailed mechanical specs would be nice.

Until HF responds definitely, the "heatsink" is a waterblock.  IIRC it is an off the shelf self cointained waterblock, pump radiator (3x120mm) assembly for cooling Intel CPU so any similar block with matching mounting hole layout should work.   The chip package is 45mm x 45mm so the key thing would simply be a waterblock with the same mounting holes.  

Data is through usb cable.  cypherdoc has stated the boards can be daisy chained   host ----- USB Cable ----->  Board A --- USB CABLE ---> Board B .... etc.  I don't believe HF has confirmed this but it seems likely.  Off the shelf USB hubs are also an option.

I agree it would be nice to have a spec sheet with various details  

Not sure if your electricity comes hand delivered in gold plated batteries or you are just really bad at math.

I am using a couple of those to produce a 4.8 KW (not a typo ) ASIC powersupply although I will probably only load them 75% to gain redundancy.  My custom prototype connector boards should be arriving next week.

Wires aren't soldered.  You (or someone) will need to buy a ATX 24 pin housing and 24 Molex crimp pins.  You would also need a crimper.   Essentially you crimp a connector onto the end of each wire and then insert the crimped wire into the ATX 24 pin housing.  You will need to do some research and find out what pins are being used by the board. Obviously if they aren't using all the pins there is no need to crimp and assemble full 24 wire connectors.   On the other end you just need to bundle all the 5V connectors together and crimp it into a ring terminal to attach to the PSU.  The same thing will all the ground wires.   The PS-ON pin can just be connected to ground to produce a connector which is always on when connected (no motherboard power switch).  A lot will depend on exactly how is the ASIC board wired.   It isn't "tough" but it does require a little practice plus it is expensive to do just one connector because you will need a crimp tool.   1 tool for 100 connectors is going to be cheaper then 1 tool for 1 connector.

I didn't realize in the first post that they would be using the ATX24 connector.  Given that the only easy solution is to find a ATX power supply with sufficient 5V current.   Still I have to say running high current on 5V rail is just a pointless complication for purchasers.  There is absolutely no reason to not use 12V.  It can be done but the buyers will face additional complication and cost for lower efficiency.  A lose-lose.

Well the ASIC chips themselves can be much much cheaper than $5/GH especially a design with high die efficiency (GH/mm2), sub $1 is certainly possible.  Still the point is that while raw chip prices will fall rapidly from $50 per GH/s down to $1 per GH/s the balance of the system won't fall as rapidly.

Lets just look at a generic large ASIC chip design.  Say 150 GH/s per chip at 100W (@ chip) with 4 modules per rig.   If the chip is running at 0.8V nominal you are looking at 125A of DC regulators.  The prebuilt modules used by KNC are about $0.50 per A in bulk.  Lets say someone went with a custom design at half that cost.   125A @ $0.10 per A =  $12.50 per board or 8.3 cent per GH.

If you use an off the shelf CPU cooler heatsink & fan assembly you are looking at $20 ea for something which can handle 150W+.   Throw in another 4 case fans at $7 ea and cooling is $108 per system or $0.18 per GH.

The rest of the board is relatively cheap.  Rather than go through ever component lets just budget $10 for all connectors, caps, resistors, fan headers, etc.  KNC boards are ~36 sq inches.  PCB production in small batches tends to be $0.20 per sq in.  We are looking at $7 for production.   Assembly and testing is going to vary by the number and type of components but even in volume on simple boards is going to be a couple bucks.  As a guesstimate lets say another $3.  So we are looking at on the order of $0.07 ea for balance of components and production/assembly.

A good quality and efficiency ATX PSU is going to be on the order of $0.16 per W.  If we assume each of 4 chips use 100W ea, 11W per board is lost in conversion, and then 8 fans (4 heatsink + 4 host) use another 6W each, the host is likely light at 10W max.   Total wattage is something like 500W.  Power ends up being $0.14 per GH.

You probably need some sort of data and power connector.  Lets say $5 ea or $0.07 per GH/s.  Finally you will need some host.  BBB is a good option cheap and yet relatively powerful and ease to develop for.  Now KNC uses a seperate expensive daughterboard with FPGA "manager" but that likely isn't necessary.  It may be possible to do it with just a BBB (or rPi) to save cost.  Say $50 per system or $0.08 per GH.

So even if magically a pile of free chips landed in your lap with no labor costs, no customer service, no profit, and no yield losses at a min we are looking at.

DC regulators = $0.08 per GH
Balance of components = $0.07 per GH
PCB production &  assembly = $0.07 per GH
Cooling = $0.18 per GH
Power = $0.14 per GH
Cables = $0.07 per GH
Host = $0.08 per GH
Total (at least) = ~$0.70 per GH.

Nobody is going to be selling 150 GH/s boards for $50.    As the margins on chips decline the "balance of the system" will act like a drag on marginal cost.  Also the prices above are rather optimistic.  More like what could be done instead of what is being done.  If you need water cooling, or want rackmount case your cost will be higher.   It also assumes a relatively efficient chip, lower efficiency means more power and cooling related costs.  Still I think chips can break $1 GH/s and the bulk of the system cost being another $0.70 to $1.00.  Even with labor cost and a small profit margin we "could" be seeing miners at around $2 per GH in 2014 but those dreaming of the day when $0.25 per GH rigs are right around the corner will be waiting a while.

That is nonsense. 

You seem to not get this concept.

The "AS" in ASIC is APPLICATION SPECIFIC.

The specific application is a double SHA-256 hash of a block header and incrementing nonce.  An ASIC can't do anything else.  It can't point GPU in the right direction, it can't do part of Scypt.  Hell it can't even hash SHA-256 in a different manner for example a triple SHA-256 hash or a double SHA-256 hash of a different sized blocked header.  It can ONLY do one specific thing.   All the SHA-256 chips ever produced and ever will be produced combined have 0.0 hashes per second on a Scrypt chain.

KNC "offered" refunds and then refused to honor that agreement in the end.

Or not.  The DC regulators on the board cost more than that. 

Can ASIC prices come down?  Yes but even with free ASICs the balance of the system is still going to have some cost and those costs are unlikely to get massively reduced over time.   An  x00 watt PSU is going to have a non zero cost.  PCB creation and assembly, DC regulators, even mundane things like fans, heatsinks, cables, and cases.   There is no Moore's law or massive markup on these components.  With free ASICs you still wouldn't have $50 boards.

No the opposite is actually true.   Symmetric encryption (like AES) and cryptographic hashing algorithms (like SHA-2) are not significantly effected by quantum computing.   Grover's algorithm makes then faster than brute force but not fast enough to be useful.   The vulnerabilities lies in public key cryptography (ECDSA, RSA, etc).

You likely won't need to increase the number of power supplies.  Granted we haven't seen the final hardware but everything about HF statements is that they are going for "off the shelf" for everything but the "mining module".   I see no reason why you couldn't make a custom build with a large power supply (say 1600W) that connects 5 mining modules to a single host.

We are all assumming a rPi can't handle 3+ modules but lets continue to assume that is true.  There are other options.   A beagleboard runs about $50 (Beagle Bone Black) and can be viewed as a "super" Pi.  It is a step above the rPi in capabilities.  Has roughly 3x the processing power, more IO capabilities, newer ARM architecture.  If (and I stress this is just an IF) a single rPi can't handle more than 3 modules I would guesstimate that a BBB would be capable of 2x to 3x that.

For systems beyond 5 total modules IMHO it likely is better to go with a more powerful host connected to multiple mining rigs.  Put 3 or 4 modules per chassis and then connect multiple chassis to a single "powerful" (i.e. low power PC) controller.

The nice thing about using off the shelf parts is it will be possible for users who want to, to come up with innovative solutions.

Well if it does work this way the advantage is ease of deployment for smaller scale systems.  For larger systems using a hub likely makes more sense.  Seems likely either would work if the host to module interface is just plain USB.  For larger systems it reduces the number of hubs.

The sierra has 3 modules per rack.  The rack is 2U high.  In theory 20 Sierras could fit in a single datacenter rack.  Thats 60 modules.   Optimally all of them will be connected to a single host to ease administration and management.  So you could run 3 USB drops to each chassis but that is a waste and since nobody makes 60 port USB hubs you are looking at multiple hubs anyways.   Another option would be to put a single 4 port hub inside each Sierra but you no have the same single point of failure.  Instead you daisy chain the 3 modules inside the Sierra together.  Then one usb drop per rack gives you connectivity for all the modules.  20 sierras + single PC/server + 20 port usb hub = 24 TH/s.

Since there is a realistic limit on how many modules can be used in a self contained (1 host, 1 chassis, 1 power supply) system I don't see daisy chain OPTION as being anything but good news.  I mean if the modules use 250W @ ~1VDC and the modules DC to DC supply is 90% efficient we are looking at ~280W @ 12VDC per mining board.   Thats 1400W for 5 modules plus the power for host, fans, and pump.  Cooling and powering 1400W in a single chassis will be a challenge.  It is unlikely one will want to try for more than that.  Beyond 5 modules it starts making sense to look into multiple "dumb" chassis connected to a mining server.

Of course IF they can be daisy chained then they certainly can be used with hubs as well.

Remember we aren't talking about USB Block Eruptor scale here.

Each module is 400 GH/s nominal.   If (and this is just an if) you are limited to only 5 daisy chained that is "only" 2 TH/s per root hub.  If you are using a PC as your host they often have more than one root hub.  In windows the device manager will show the number of root hubs.  My PC has 4 which is pretty common.  That would be 4x the daisy chain length. 

If you need more then use a hub.  Even if you are limited to less than 127 remember we are talking about 400 GH/s nominal.   If you are limited to say 60 units effectively per host (not a rPi but a real entry level PC/server) that is "only" 24 TH/s.  That is going to fill an entire server rack.  I think people looking to deploy multiples of 24 TH/s can afford one host per rack of mining hardware.

It doesn't work that way.  x 512 qbit QC can't do anything more complex than 1 512 qbit QC although they can do 2x as many simulations.  


As posted many times D-Wave is not a general purpose quantum computer.  DWave is a type of specailized quantum computer which uses a concept called quantum annealing to simplify complex simulations.

http://en.wikipedia.org/wiki/Quantum_annealing

DWave is not capable of and was never designed to implement Shor's algorithm. It has absolutely no ability to break ECDSA keys.  Even if D-Wave was a general purpose quantum computer and capable of implementing Shor's algorithm against ECDSA keys it would need to be on the order of tens of thousands qbits.  Lastly an address is the hash of the public key.  Even if a general purpose quantum computers with sufficient qbits and capable of implementing Shor's algorithm against ECDSA keys the PUBLIC KEY is still needed.  If coins haven't been spent from an address (no address reuse) then the public key remains unknown and thus unsolvable.

The % of the MPP is already in the terms.   They are just clarifying that they will provide full boards not just raw chips.  No racks are included for Sierra or Babyjet, neither are additional power supplies. 

A BabyJet consists of 1 mining module.  A Sierra consists of 3 mining modules.  If due to the MPP you receive 400% more hashpower you would receive 4 mining modules per BabyJet order, and 12 mining modules per Sierra order

Well it has been a week later and multiple people have asked both on the forum and indicated they sent emails with absolute silence coming back from the company.