Re:Inside Telegram’s ambitious $1.2B ICO to create the next ...
by tudorw on Tuesday January 16, 2018 @05:41AM
(#16157192)
While the economies of scale often offer a discount, there are some possibilities here, power cost increases when demand at a single location increases beyond a certain factor, inherent network capabilities suit a decentralized service, network services can route around censorship and handling heat and safety cost are not linear when resources are centralized.
So perhaps switch the laptop for a wall based crypto-heater packed with silicon (see here for the heat output of a gaming pc vs a space heater https://www.pugetsystems.com/labs/articles/Gaming-PC-vs-Spac...) and there may be some potential there, maybe Elon can throw a battery in there too...
It's certainly fascinating that for all the talk about blockchains and ICO's I've read very little about the fact that the internet has been augmented by a distributed cryptographic system available for rent, altcoins that rely more on CPU and memory and therefore discourage GPU and ASIC use might mean the network moves to more general computing power for rent, interesting times as they say.
Re:Inside Telegram’s ambitious $1.2B ICO to create the next ...
by jacques_chester on Tuesday January 16, 2018 @07:02AM
(#16157560)
Altcoins rely on economics. If you're a miner, you invest your fiat in mining coins showing sufficient value and liquidity. If an altcoin rises high enough, ASICs will follow.
I've seen this interaction so many times now. It's great that people read these whitepapers, but what to they judge them on? They're often a one-sided technical explanation with zero objective information in terms of feasibility in the market.
There's one (apparently delayed) ICO planned, for a company that wants people to share their CPU time to train neural networks in exchange for some token. I can hardly believe that a distributed network of laptops in living rooms can be more efficient (and therefore cheaper) at doing this than centralized, purpose built hardware in data centers with efficient cooling systems (like Google's new TPU). And that's even if they figure out how to make this technology work effectively.
My colleague, though, was very impressed though, because he heard somebody say that training a neural network takes a lot of time, which is true, depending on the complexity and whether you run it on a €500 Acer laptop or on specialized hardware.