20 million dollars have just been invested in BitFury (July 2015), totaling 60 million which this company has raised (source: coindesk). There was nothing like this for at least 5 months.

On Mining Profitability

I find it very surprising that people invest in bitcoin mining. Why?

My [private] conjecture is as follows:
mining profitability is almost always negative.

More precisely [in a market equilibrium situation] given the anonymity services mining provides, people are willing to mine at a loss. This is basically because freshly mined coins have no origin and cannot be linked to the origin of funds used in their production.

Having said this, negative profitability property holds on average, and for a substantial proportion of miners. It could be positive for people who are able to produce better ASICs than their competitors. And have been even more negative for people who trusted companies such as Bitmine.CH to deliver miners for them. In fact IMHO all miner companies in existence have deceived their customers in some way, but some of them have deceived them a lot more… see also Section 2.4 in this paper.

Some Works on Technical Aspects of Mining

Mining with SHA256 calls for many very special optimizations which allow to reduce the IC cost and energy consumption:

• There are countless highly technical papers about implementing SHA256 with traditional techniques such as CSA adders, see bibliography inside our paper.
• However many optimizations were NOT covered by traditional SHA256 ASICs developers, because there are also many improvements which are SPECIFIC to bitcoin: see sections 10-12 and pages 16-24 in our paper.  This is about saving some 37% w.r.t a naive implementation. Savings are obtained in the first and last rounds of the computations.
• Here is a shorter conference paper about these optimizations, cf. this paper as published by Springer.
• More advanced optimizations for additions (carry reduced adders) are explained in this blog post, expected to save another 5%, maybe.
• Another family of interesting optimizations are those which aim at saving the message expansion in the second application of the SHA256 compression function: with several specialized engines with a fixed nonce and variable mid-state H0, see page 118 here. Allows to save some 25% for one out of three compression functions. There are some technical issues with making sure that there is enough variability in the first 512 bits, without re-computing the Merkle Root too frequently.

In other works we have also studied the Stratum protocol used in mining, cf. this paper and this blog entry.