Speaking as a cynic, I have to admit that whilst I think  every new company has to start somewhere, jumping into your first asic design with 20nm seems like technical and commercial suicide. It's still essentially a 'development' process with an awful lot of kinks to iron out. The NRE is horrendous, as is the wafer costs, and a small company is always going to go to the back of the wafer queue.

Aside from the very thin technical detail, which I would have thought would have been very prominent in the FAQ's, the power 'specification' of 1100 watts for 5TH seems rather suspect. 1100 watts at the wall translates to about 80% of that 'at the chip' as it goes through 2 voltage conversions, the first from 110/230v AC to 12-20V DC, then down to 0.7 or so volts. So 1100W for 5TH translates to an overall rating of 0.22W/TH (or 0.18W/TH at the chip)

Now that's not impossible to achieve, but not with a standard SHA256 gate based design, and since they don't seem to have any previous asic design experience I doubt they would try anything fancy. So the figure that would seem likely for this design would be 0.4W/GH at the chip, a factor of 2 times higher.

Perhaps they might want to explain how they have achieved what they have claimed. I know of a few answers that are valid, so I'm interested to hear what their response is.

Here's a little equation about power to consider:

Watts/Hash = Ng * Pg * F / (Nc * F) where:

Ng = number of gates switching per clock cycle - a design constant which depends upon the pipeline stage architecture
Pg = average switching power per gate per Mhz - a silicon process constant; about 0.4 nanowatts per Mhz for current 20nm processes
F = clock frequency (variable)
Nc = number of cores (pipelines) in the device; each core produces one result (hash) out of the pipeline every clock cycle, the pipeline latency is
       ignored as it's irrelevant in practice. So hashes = NC * F.

Or to simplify, P = Ng*Pg/Nc.  F cancels out, ie frequency is irrelevant. Static device power is also ignored here as it's relatively low in comparison.

Well said. But I'd be a lot more impressed if there was a real 'flow' of information rather than a snippet with nothing to back it up. These are digital chips, not mixed signal where there's a lot more issues. If the chips had a built in test structure it would be very easy to run specific test patterns to see what the problems might be. This isn't a new CPU core - it's a bog standard fixed integer arithmetic function that can be realised in less than 5k gates per stage (or any awful lot less if you're clever enough) then simply replicated many times to build the pipelines. So where's the problem?

None of my business, of course, simply professional curiousity. Contrary to what some of you may think, I do wish Cointerra well and hope that they live up to their customer's expectations.

I'm only concerned that paid up customers get told the truth about what's going on. If a company is ahead of it's delivery date, then it's nice to be kept informed. When you're late, you need to communicate any potential delay and/or problem right away, preferably with a fix.

Appreciated, amigo. I've been 'there' before many times in my professional life and this situation is unfortunately following a classic path, engineers tend to know within the first few hours if they've got a dog. Don't get me wrong, I want their project to succeed and their customers to get what they ordered, but if this was a commercial company they were developing a system for, the customer would be on their doorstep and in their labs wanting answers.

Just had a visit to Cointerra site where they posted, on January 14th,  that they had a fully built up box working at 1.62TH, or " a little over 80% of (their) performance target.

They also said that their engineers were "making continuous tweaks and refinements each day to improve the power efficiencies on the board to increase performance". Well, they said the engineers had one die on the package running at 132Gh/sec on the 7th of January, so four of them should meet the target spec. They would have known within hours if there was any kind of problem, which there clearly is.

Maybe I'm just cynical (and I am) but 20% seems an awfully large gap to close with 'tweaks' unless someone has seriously screwed up on a major part of the design. I haven't bought one of their rigs, but I'm sure that every one of their anxious customers would gladly sacrifice a bit more power consumption just to get their paws on their purchases. Remember that Cointerra have $1.5 million of Private Equity funding - which they obtained before the public got in on the act - and I'd hate to think that any delay had been introduced into the supply of rigs to the public customers as a consequence of the presence of this funding.

Remember that these guys are already running late and if there's still 'tweaks' to be done at this stage, the December buyers will be lucky to see their rigs before the end of January; small 'tweaks' might mean changes to printed circuit boards or different components to be sourced, either of which might be $2-3 changes but may take weeks to source and/or change.

Perhaps one or two of those customers might like to ask Cointerra some very pertinent questions about what the problems are? Don't want another Hashfast, do we?

Yes, delivered by airborne porcine quadrupeds.

The real answer is that no one knows what will happen. 18000 million is probably a bit high, anyone seriously mining just know knows that a ton of new kit is going to get dumped on the network starting mid February when all the chip orders from November start to come through to production. 1800 million would very roughly equate to a total network hashng power of 130,000 TH; from the present 14,000 TH that's adding 116,000 TH, or nearly 60 THOUSAND 2TH rigs. None of the current or 'pending' suppliers have production capabilities for units of this size and complexity anywhere near 1000 a week at present, and it would take them quite some time to get up the learning curve. The 'hidden' mining companies are another matter altogether, but I'll bet even they can't turn on instant capacity via their subcontractors.

So buyers will tread carefully unless they get offered a stellar deal - or the value of BTC jumps again which will start a whole new cycle.

But I'd also advise anyone thinking of buying to consider this - if you were an existing miner earning good money on your existing setup, would you want to encourage new entrants who will push the difficulty up? So be careful of what you read, there are hidden agendas!

Your right about the electricity cost, it's being conveniently ignored in most cases. If you believe the rig suppliers, 1TH is going to need about 0.7kw of power to drive it. At UK costs that's close on 20 cents an hour, $4.80 a day and $144 a month. But I reckon by the end of 2014 that the same TH will earn a lowly $400 a month ..... so at least a third of earnings will go on power.

I beg to differ, there is a minimum cost you can manufacture an asic based miner for (in volume), and it's easy enough to calculate if you have commercial design experience or are willing to do a reasonable amount of research work on the net. Granted, some companies have more 'overheads' than others, but that's just greed or poor management. I'm talking about a well run company with experienced people and modern manufacturing methods.

Very competitive, but the chances of that happening are zero. It costs about $0.9/GH to actually build rigs, so if the difficulty keeps going up and bitcoin value stays the same there's a natural limit as to when it simply not worthwhile for a company to make and sell them - unless they're mining with them, of course, but then why would they want to subsidise their paying customers?

The big problem with SHA256 engines is their power dissipation. It's a consequence of the architecture that means that nearly all of the engine is active nearly all of the time, unlike an Intel i7 where it's really only the ALU that has a heavy load. Going from 28nm to 20nm saves about 40% of the relative power to do the same function, OR, with the same die size give you about 50% more hashes per second. There is no game changing technology on the horizon that will change this, ie one that would give you a 10x or even 5x improvement.

Also, 20nm is still effectively in development, so wafer and die yields may not be too good. But if you've got $20m+ of other people's paid up front money, I guess you can take the chance and do it as a vanity project, even though it's not necessary. 28nm is more mainstream, but even then it's not without it's problems, but the NRE is a lot less than 20nm. A well designed 28nm engine will easily beat an inefficient 20nm one, bear that in mind. 

You know the old saying: greed and a poor grasp of basic arithmetic makes people do strange things.

I'm sorry to say it, but I don't think you're going to hear anything else from them. They've essentially given no information out for 2 weeks now, and that alone should tell any customers that all is not well. It may be time to get legal with them, I know they're an LLC but the directors still have a Duty of Care to operate the business responsibly, and can be jailed (in the UK at any rate and I'm sure the US is the same) if there has been gross mismanagement or abstraction of excessive funds from the company's coffers.

I still hope they will 'man up', talk to their customers and agree a recovery plan with them.

When did you llast see a quotation or invoice from TSMC or Global Foundiries for 28nm?

I like this thread, it's good work and an excellent community resource. I hate to be a party pooper, but your wafer prices are way too low by a large margin. Qualcomm might get an untested 28nm wafer for $3500, but no rig vendor is getting their 28nm wafers any less than about $20k a pop - wafer fabs can sell more 28nm than they can make just now, the Bitcoin asic business is simply background noise to them.

That makes a whole lot of sense.

There are several possible explanations for the Hashfast situation:

They have run into mammoth technical difficulties which they don't know how to solve
Their attempts at making an asic went wrong and they've run out of cash to try again
They had no intention of making rigs and have stolen customers money
They paid themselves too much and underestimated production costs for customer systems

Feel free to add your own.

I have no interest in Hashfast save that I don't like seeing hard working, trusting people getting a raw deal from suppliers. If they are genuine, they could at least try solve their present situation by being honest and communicating with their customers.

Lots of complex technical projects go wrong, the trick is to keep customers in the loop and tell them whats happening, even if it's bad news. An information vacuum very quickly gets filled by speculation. If Hashfast have had some kind of spectacular failure, I'd urge them to at least come clean and tell their customers what the real situation is. People are a lot more understanding if you at least give them the chance to understand the problem, at this stage they might still be willing to wait and help work out a recovery plan if they have some chance of
actually getting their rigs.

Well, it's easy to see why people will think that, but all is not what it seems. If the difficulty goes up by a factor of 10 by July, then the network will have an extra 120,000 TH on it. That's 40 THOUSAND 3Th boxes, just think about that for a moment. IF, and it's a big if, vendors do manage to get 500TH chips made in volume and, crucially, yielding decently then to manufacture that volume of boxes at 120 die per wafer requires 240,000 chips / 2000 wafers. (or 40 25 wafer batches). That's a close on $40 million investment in chips alone, and for what return?

Well, at that stage a 3TH box will earn around $2500 a month. That's fine if you've paid 2$/GH because you might get to break even in maybe 3 - 4 months. If you paid $3/GH then it might take around 6, depending on how much investment is put into the network after you get your box.

Considering the amount of difficulty rig vendors seem to have getting hundreds of units out of the door (or any at all, in some cases), it will be interesting to see how they fare with large volumes. The real problem lies with the dedicated mining 'companies' who presumably have their own chips, but even they won't be keen on an ROI of possibly less than 50% a year.

Naturally, if your going to spend a good few thousand dollars on a rig, you want some reassurance that you will at least get it back in a reasonable length of time. No one can give you that at present, but if you have modest expectations of profit then the outlook maybe isn't so bad.

Well, a lot of people clearly believe that it is - look at how much money KNC took for their (overpriced) 20nm machine. You can do the sums yourself - just extrapolate the difficulty forward to April, May etc and then you can make your own estimate of profitability. From various figures I've seen, you might just get your money back in 5 - 6 months, after that is anyone's guess.

Of course, there could be a major rise in Bitcoin's value as more and more people adopt it?

This 'advice' really doesn't help you much, does it? Ultimately it's you that has to make the call.

Think again. When the Bitcoin value exploded in early November, every mining company and probably a good few rig vendors were ordering 28nm wafers like crazy - some $40 million worth, from what I've been told, corresponding to about 110,000 TH which will be added in February/March - it takes about 11-12 weeks to get the parts plus 6 -8 months to assemble.

Not good news, sorry.