5 ms·
So I looked into doing this on a personal level a while back and did some back of the envelope calculations and it was hard to square the economics of it. Two
by shalmanese 11y ago
So I looked into doing this on a personal level a while back and did some back of the envelope calculations and it was hard to square the economics of it.
Two factors made it really difficult:
1. All computation is done in DC. Even if you assume all other prices fall to 0, the AC->DC conversion cost on a per watt basis makes it hard to compete with cheap heaters which can operate on pure AC. Just power supplies capable of powering 1000W of computation costs a lot more than a 1000W heater, even used. You need to get the economics exactly right to make this scheme profitable and the factors are so sensitive to shifts that it's hard to plan for the capex.
2. There's a noise/convenience/space tradeoff that's hard to square. Ideally, you want the heat to be close to the users but the noise to be far away. It's hard to make electronics perform well above 100C as opposed to resistive elements that work fine up to 1000C. That means either moving air quickly (noise) or moving it over a large area (space). Neither of these are ideal.
It's possible for these two factors to be overcome but they're sizable and makes me uncertain whether it'll ever be possible to scale this as a viable business vs a cool idea that gets trotted out in trend blogs once every few months.
- jacquesm 11y agoThe whole trick about any kind of 'co-generation' is that you get something you need (computation) and something that you can use that would otherwise go to waste (heat). The only time that scheme does not work is if you don't need the heat.
- shalmanese 11y agoRight, but the profit from co-generation from 100W of computation has to be at least greater than (cost of generating 100W of DC) - (cost of providing 100W of pure heat). I was looking at it in the context of bitcoin mining at the time but it's remarkably hard to find computational tasks that both generate good $/FLOP returns and are stable and scalable which is what a scheme like this needs.
- tinco 11y agoI'm a bit confused, don't the losses of an AC->DC converter go into heat as well? Why is that less efficient than a cheap AC heater? Is the energy lost in some other way or is the heat generated by the converter not usable?
- jacquesm 11y agoYou're not confused at all. The computational device is a 100% efficient space heater and at the same time does something other than just heat space.
- shalmanese 11y agoI'm not talking about energy, I'm talking about $
- CarVac 11y agoNo, the profit of co-generation from 100W of computation has to be greater than (the productivity loss from the geographic dispersion of the computational resources minus the cost of actively cooling the computers). The electricity is free because it would already have been spent on computation (plus some more on cooling that computation, which is no longer necessary).
- reacweb 11y agoWhat about the energy conservation. Of course there are energy leaks in AC->DC conversion if you don't take into account the heat produced during the conversion. If your purpose is heat, the efficiency is 100%. A radiator behaves like a very big passive cooler for your data center. You want it to be less than 60°C to avoid getting burned. That is perfect for CPU. Fans are definitely far from mandatory.
- cosmolev 11y agoWe actually have to compare with heat pumps. They provide 3KW of heat per 1KW of AC electricity consumed.
- Retric 11y agoDepends on outside temperature, if it's cold enough heat pumps become less efficient than simple resistance heating. The transition point where they become useless is generally around 0F which is not all that cold. Granted if you live in a warm area then it's generally not an issue.