Designing my own RPG, and transferring human consciousness into machines is a part of the background. Wondering what are the real world numbers for doing such a thing. Memory, processing power, heat generation, any details that would make it more plausible.

Thanx in advance.

  • $\begingroup$ Welcome to WorldBuilding Jason! It seems like your question has already been answered. Please take a look at the proposed duplicate. If you think that the linked question and the answers there don't sufficiently answer your question please edit your question to explain what differences need to be discussed. Meanwhile I am voting to temporarily put this on hold until you edited it. If you have a moment please take the tour and visit the help center to learn more about the site. Have fun! $\endgroup$ – Secespitus Mar 18 '18 at 11:25
  • $\begingroup$ is the transfering for virtual-reality style uses or something else? $\endgroup$ – Ajnatorix Zersolar Mar 18 '18 at 11:43

A human brain is thought to contain about 86 billion neurons and these neurons can be modeled reasonably accurately bu a set of four equations known as the Hodgkin-Huxley model, which has four variables and is first order, meaning we need about eight numbers to store it's state at any instant.

So assuming each number can be stored as a standard double precision floating point numbers ( 64 bits each ) then we need $4.4\times 10^{13}$ bits or about 5.5 GB.

Which is a surprisingly small number, well withing the reach of current computers.

Of course doing 86 billions calculations involving solving 4 simultaneous differential equations with any reasonable speed is well beyond anything we can currently do in a brain-sized object.

Apparently the peak rate a neuron fires is about 1000 Hz and that's the minimum to consider a computer brain "simulation". That means we need something like to be able to do about $9\times 10^{13}$ of these mathematical solutions per second, and each of those would take perhaps a thousand floating point operations. So that's ballpark $10^{16}$ Flops.

That's about the same computing power as IBM's Sequoia super computer, which occupies about $280\,m^2$ and consumes a staggering 6 megawatts of power, all of which has to be dealt with for cooling purposes.

Note that the 5.5 GB of memory can't be standard memory like your PC has. We'd really be using memory-per-neuron with some working store. More like a giant GPU with lots of local memory for neuron state and local calculations and access to global memory when needed. That would considerably increase power consumption and hence heat dissipation issues. At least an order of magnitude.

  • 1
    $\begingroup$ You forgot to compute the memory space required to store the state of the links between the neurons... $\endgroup$ – AlexP Mar 18 '18 at 19:46
  • $\begingroup$ @AlexP Well spotted. If I get a chance I'll revise that. $\endgroup$ – StephenG Mar 18 '18 at 20:24

I read a lengthy description (I think it was James P. Hogan's Reality Interrupt) and at that time I checked the numbers, which checked out; Hogan was way down alternative-science alley, but he usually did do his research.

His "brain emulator" was based on a cupboard-sized stack of silicon chips (probably achievable now) containing around 100 billion "neuristors", each composed of a tiny CPU and a memory of two kilobit - or was it kilobytes?. The chip was actually a latticework through which a dense cooling fluid was continuously circulated.

The power requirement was in the hundreds of kilowatts range, so the actual machinery, with heat pumps, fans etc., was way larger than a head - more like a large room.

Comparing these numbers with StephenG's answer, they aren't that far away, and Hogan's were probably based on the same assumptions.

Solving an ordinary differential equation of this kind, with dedicated hardware, seems way less costly than StephenG's estimates by almost three orders of magnitude.

Also, it stands to reason that a sequence of neural operations would not be completely unrelated, but instead (very probably) linked together, so that there would be ways of optimizing results and saving time.

A major difference lies in neuron firing frequency - while true that a neuron has a maximum firing rate of one thousand per second, the average rate is way lower by almost four orders of magnitude.

So, the actual transition frequency and power requirements go well down, back into the range estimated by Hogan - around 100 Gflops without any optimizations.

It is somewhat disturbing that these numbers also happen to match the computing specification of Intel's "Movidius" deep neural processor...

(And just a few days ago, our power requirements got further reduced. We might be able to fit this in a head after all, and maybe power it from a human metabolism and a Seebeck converter).


Not the answer you're looking for? Browse other questions tagged or ask your own question.