AI vs Brain Energy Efficiency Calculator

Credit goes to Claude 3.5 Sonnet!

GPU Calculations:

Let's consider a GPU using watts and capable of TFLOPS (trillion floating-point operations per second) in FP16.

First, we convert the GPU's performance from TFLOPS to operations per second:

TFLOPS * 10^12 = ops/s

Then, we calculate operations per hour:

ops/s * 3600 s/hour = ops/hour

We convert the GPU's power consumption to kWh:

W * 1 hour / 1000 = kWh

Finally, we calculate the GPU's efficiency in ops/kWh:

ops/hour / kWh = ops/kWh

Brain Calculations:

Note: The following brain calculations are highly speculative. Expert opinions on how to quantify brain computations vary widely, and the correspondence between biological synaptic activity and artificial neural network operations is not well established. The values used here are simplifications and should be treated as rough estimates for comparative purposes only.

The brain has an average of trillion synapses. Let's assume each synapse switches times per second and computes a function that corresponds to neural-network weight. The brain consumes approximately watts of power.

We start by calculating the brain's total operations per second:

trillion synapses * switches/s * weight/switch = ops/s

Then, we calculate operations per hour:

ops/s * 3600 s/hour = ops/hour

We convert the brain's power consumption to kWh:

W * 1 hour / 1000 = kWh

Finally, we calculate the brain's efficiency in ops/kWh:

ops/hour / kWh = ops/kWh

Comparison:

The ratio of brain efficiency to GPU efficiency is:

/ =

This means the brain is performing times more operations per kWh than the GPU.