Can PoW be as energy efficient as PoS?

Lots of discussion on Twitter recently regarding PoW vs PoS. People like Jack Dorsey from Twitter arguing that PoW vs PoS is not black vs white, for instance considering solar or places in the world where the price of energy is zero (like deserts or wind tunnels)

Here is one proposal on how PoW can be made as energy efficient as PoS.

  1. Energy consumption of chips comes mostly from transistor switches (a transistor is a capacitor, and energy per switch is C V^2/2 where V is voltage and C is capacitance.

  2. Transistors essentially switch on each clock cycle, so if you need to have crypto algorithms that require fewer clock cycles and larger chip area. In this case, energy use will be small and costs of chip design and production will be large, so energy will be a small percentage of mining costs. In the extreme case, you have a huge chip with a very slow clock and a crypto algorithm requires such a chip to run efficiently.

  3. The question is, are there hardware people on this board that could help to design a chip/algorithm like that?

Energy is consumed not only when computing POW, but also in the hardware manufacturing process. If this proposal trades one kind of energy consumption for another, how does the total energy use decrease?

Hey Elbeem - I think energy consumed during manufacturing process is amortized over the life time of the chip.

So in the perfect world the chip should be produced once and then work for long long time …

1 Like

Hmm, if the total energy consumption, and thereby the total cost of a chip during its lifetime is smaller than for existing POW chips, wouldn’t that just lead to mining being more lucurative, attracting more miners until the total energy consumption is the same as today? The only difference is that there will be more chips.

Hey Elbeem

The equation is

MiningCost + ProfitMargin = BlockBounty

OR

EnergyCost + OperationsCost + ProratedDesignCost + ProratedChipCost
= BlockBounty

OR

EnergyCost = BlockBounty - ProratedDesignCost - OperationsCost
- ProratedChipCost

Therefore, you can bring energy cost to almost zero for a fixed block bounty by tuning the rest.

Agree?

I’m not quite sure what is the proposal here. There’s no point in designing chip that runs slower because the algorithm can always be emulated on more efficient processor.

This is in principle the same as acquiring ETH (large initial cost) and running a validator in PoS (low energy use).

Hey,

It is not about the chip running slowly.

Computing power of the chip is essentially clock cycles per sec multiplied by the amount of computation you do per cycle.

The point is about using algorithms that use less CPU cycles, but do more computing per cycle and use a larger computational module area on chip. These chips will be expensive to produce but cheap to run in terms of energy. So you shift the costs from energy to chip production.

In that case, what would prevent someone from raising the clock rate of the CPU?
If I had such processor, the first thing I would do to maximize profit is increase the clock rate to the point where cooling starts to be the problem. But this means it dissipates a lot of energy and we’re back at the beginning.

I don’t think this problem can be avoided as long as computational power correlates with profit.

Hey

If the processing area is large, the electric signal needs time to propagate.

So clock time needs to be larger than propagation time. The larger is the area, the more time it needs to propagate.

OP may be interested in the research around Optical Proof of Work, using a hash algorithm designed to be implemented in energy efficient photonic chips that use optical switching. See: oPoW Resources — PoWx

As identified above, this changes the energy inputs in the PoW from CAPEX + OPEX for the ASICs + electricity to primarily CAPEX for the photonic chips up front. Other than that, as far as I can tell, it retains many of the features of current generation PoW: dependence on specialized hardware (and supply chains) plus obsolescence over time requiring continuing re-capitalization to sustain security.

Are there any tasks that can only be done efficiently by humans, but can be verified efficiently by machines? If so, you could imagine a chain where PoW is done by existing humans, providing employment, rather than electricity guzzling chips.