This video is Jörn Loviscach’s talk on Bitcoin mining given at a workshop in September during the Gesellschaft für Informatik 2012 conference. He describes the environmental impact of Bitcoin.
His numbers do show how CPU mining is roughly an order of magnitude less efficient than GPU mining. He confirms how GPU mining too is rougly an order of magnitude less efficient than the technology for FPGA mining.
Presumably because there is no ASIC hardware shipping yet his analysis didn’t consider ASIC mining but ASIC hardware too is likely somewhere in the range of being an order of magnitude more efficient than its FPGA predecessor.
The challenge that Mr. Loviscach didn’t quantify was the ratio of current mining that is performed by each type of technology. Without being able to estimate that ratio, any estimates about any electrical consumption can vary (by over-estimating consumption) nearly 10X from the actual consumption.
For some of today’s miners, including many of the larger ones, nearly 100% of all hashing is performed on the more-efficient FPGA hardware. However many of those mining who pay little for electricity have been adding GPU capacity, especially since they can absorb the used GPU castoffs at a lower purchase price.
The plurality of miners are individuals running a GPU or two on an existing computer and most haven’t yet switched over to FPGA, though when ASICs start shipping that hardware may obsolete their GPUs in an instant.
An educated guess might come up with a ratio of nearly two-thirds of mining is still done with GPUs (though a small amount of that includes CPU mining yet, including that performed by botnets) and FPGAs are responsible for the remaining third of mining today.
Thus the power consumption results provided are possibly 30% too high, and half a year from now total power consumption for the entire network may drop well below current levels thanks to the one-two punch of the block reward subsidy drop plus initial ASIC shipments that could force nearly every GPU into retirement.
There was another point made in the talk that needs clarification. Mining activity is nearly the same no matter how many Bitcoin transactions are made. So more transactions do not require additional mining and thus no noticeable difference in electrical consumption.
Calculating the energy cost today on a per-transaction basis and then extrapolating that out to cover all transactions produced for all commerce simply results in some ridiculous total power consumption number that has no bearing with how the technology works.
It is like calculating the cost of electricity per passenger for a subway trip at midnight (which transports a total of nine passengers) and then use that result (e.g. $25 worth of electricity per-passenger) to calculate how much power the subway would consume during rush hour when the same subway is packed with several hundred passengers. It just doesn’t work that way. Length: 24 minutes.