For better or worse, computation is not unifiable. Each computation generally needs a certain amount of storage space, a gpu that can run N trillion floating point operations per second and K billion tris per frame, etc. You cannot pick a single number, in a reasonable way, to represent all that.
On Ethereum, executing a function on a smart contract means you're doing computation and maybe writing data, which involves both CPU and disk. The amount you pay for running this operation is denominated in gas, and is determined by things like iterations, memory used, data written, etc. The end user doesn't get all of this separately, but instead gets a unified computation price. Maybe it sounds unclear at the current scale of Ethereum (which can only perform basic operations), but at a larger scale, we could unify the cost of remote computing in a single metric.
I'm not sure where this is going. Doesn't the computational cost to mine ether bitcoin or ether change over time?
And also, we already have units for quantifying computation. We can talk about millions/trillions of floating-point operations and also millions/trillions of instructions executed. The problem is that it's very hard to reach the peak theoretical capacity of a system, and also, when you're talking about trillions of instructions, the time taken depends on your instruction makeup.
You'd get a similar problem if you talked about ether as a computational unit, because a machine can be designed to mine ether really fast, but not be suitable/optimized for other types of computation, meaning that the compute power it takes your system to compute an ether might not be a good way to quantify the time it would take said system to compute something else.
The idea is to use Ether as a unified and simple unit for computation cost in a sense that it can become a recognizable metric. Just like instinctively thinking "this is cheap" when you see something priced at $0.01, you could think "I can run this" when you see something priced at 0.0001ETH.
Perhaps it’s important to remind ourselves of the purposes of Ethereum and Bitcoin transaction fees. Bitcoin stores state in UTXOs, so transaction fees are paid for block space, and scrypt removed the loops and recursions to prevent a DOS attack. Ethereum is a memory state machine, so the fee is mostly for the expansion of that state, and additionally enables loops and recursion while using gas as DOS protection. In both cases, the actual processing cost is hardly relevant, and poorly implemented. I don’t doubt the utility of a blockchain based on computationally-relevant gas, but that would probably be a different type of project.
First of all, ethereum computations are ran redundantly on all full nodes. This means the cpu power of the network is in kHz or less.
Secondly, computation is a proxy for energy consumption, which is how most crypto mining costs correlate with. Every electronic device in your home is simply a space heater without a fan. It just happens that the devices do useful work while heating up the circuit wires in them.
Clearly the author put effort into making it brief and easy to understand. In the distributed network of computers participating in ETH, one can associate price with computing potential. For a given finite calculation, one can ask how much it would "cost" in ETH, also giving one a reasonable metric for how much modern-moment computational power is required from the ETH network. Maybe the network is always growing, so maybe this is variable, but maybe some computations are variable and grow parallel to the growth of the ETH network, so one could track that subset of calculations or calculations-required in terms of ETH. It's a nifty idea.
> For a given finite calculation, one can ask how much it would "cost" in ETH, also giving one a reasonable metric for how much modern-moment computational power is required from the ETH network