U.S. engineers come up with a way to increase the profitability of mining bitcoins by 30%
Dr. Rakesh Kumar from the University of Illinois at Urbana-Champaign in his work “ Bitcoin mining with approximation ” described a mechanism by which you can increase the profit from mining bitcoins by 30% - in terms of unit of processor power spent. The method is based on the use of approximate calculations - if a small number of errors are allowed to appear, the resulting calculation efficiency increases.
Approximate calculations are often used in science to simplify mathematical calculations. Dr. Kumar calculated that using iron, which counts with a certain, but not 100% accuracy, it is possible to increase mining efficiency by 30%.
When mining bitcoins, it is necessary to calculate hashes from certain initial values. It is necessary to obtain a hash of a certain type, and since it is impossible to predict the type of hash before its calculation, it is necessary to carry out calculations, and thereby prove that the processor time has been spent.
Due to the constant planned increase in the complexity of bitcoin calculations, miners gradually switched from CPU to GPU, FPGA or ASIC. An increase in complexity requires an increase in processor power consumption. Kumar and his students thought - is it possible to understand mining technology and reduce energy consumption, increasing profits?
As a result, they turned to the idea of approximate calculations. According to Kumar, quite a few applications running on a computer - for example, visualization, -
can work with calculations that are not 100% accurate. Such programs are error-resistant and can work easily even on inaccurate hardware. Researchers have found that mining bitcoins has the same properties and “forgives” some inaccuracies.
The researchers found that the processor, which is required to guarantee 100% accuracy of operations, in some cases consumes 2 times more energy than the processor, which must produce 99% accuracy.
Bitcoin mining can be performed in several parallel processes, and the iron used for this consists of independent modules. If one of the modules makes a mistake, it does not affect the work of others.
As Kumar explains, mining errors can be of two types - the right decision, which was mistakenly considered incorrect (false-negative), and the wrong, mistaken mistake for the correct (false-positive). Since the probability of finding the right solution is very small, false-positive can be neglected - they will be few. And they can still be sent to the network anyway - all the same, other miners will check it and filter them out when updating the block chain. And a false negative decision is a missed opportunity - the right decision that has not been implemented.
Dr. Kumar argues that using iron, which counts hashes approximately, will allow you to pack more modules in the same space than regular iron. As a result, the time to calculate the hash is halved - the miner can generate twice as many hashes per unit time. Due to periodic errors, real performance increases by about 30%.
Based on his work, Dr. Kumar will make a report in June 2016 at a conference on electronics and automation design.