Naked Science Forum
Non Life Sciences => Technology => Topic started by: Alasuya on 17/08/2013 02:24:38

Hello, I have a few questions regarding computational power. In the far off future, I imagine creating a realistic virtual reality, a cybernated solar system composed of a sun like star and numerous habitable planets and moons with the same complexity of life on Earth. For this solar system I would like to create 100,000 computed humans in it. All of them would have consciousness. And perhaps the ability to even download my own consciousness into these virtual worlds.
But there are issues with this concept, the first one is no one knows exactly why we are conscious beings, that is currently being researched, the second being is processing power, all of this simulation would require huge levels of computation power. IBM researchers estimate that even just one human brain considering it's complexity would compute at 36.8 petaflops, or 36.8 quadrillion of data per sec. Other researchers with microtubules estimate to be closer to 10^28.
What is the theoretical max limit of computing power for a computer one kilogram?
Seth Lloyd calculates an upper bound for a 1 kg computer of 5*10^50 logical operations per second carried out on ~10^31 bits.
Are his calculations accurate to what the true upper limit is?

Mass is not the only constraint on computing power  there is also a theoretical minimum energy for a logic state change. The faster you change logic states, the more power you consume. It is no good to have a computer which has a mass of 1 kg, but which melts into a puddle as soon as you give it a significant problem to solve.
The question is also a bit vague in terms of a traditional binary computer vs a (currently somewhat theoretical) quantum computer. There are some types of problems for which a quantum computer with just 256 qubits would theoretically have more processing power than a computer with the mass of the Earth. Whether simulating a solar system (or a brain) works better on a quantum computer than a traditional computer architecture requires answers to a number of currentlyunsolved problems.
With current computer technology, simulating even a small fraction of a brain consumes far more mass, power and volume than a human brain. It is not currently known which parts of the simulation model are significant, and which are not. Once the significant components of neural connections are identified by use of software models, it should be possible to build hardware neural simulators which are far more compact and powerefficient than the software models running on generalpurpose computers.
There is also the question of the fidelity of the simulation model. At one extreme, it is possible to model the Earth as a point mass at a certain distance from the Sun, having a certain velocity. However, if you want to model the Earth down to the level of trees, microorganisms and whole ecosystems, that is an entirely different category of simulation model, requiring an entirely different category of computing power!

Thanks for the response!