Traditional datacenters usually came in one of two general forms, Colocation datacenters where anyone can rent rack space in the building, and your hyperscaler datacenters like AWS and Azure.
Colocation Datacenters are usually only partially full of servers at any given time, and are often rented by private enterprises running small time web applications or even just internal tools.
The hyperscalers are worse, but the business model is overbuilding capacity but not all the compute is actually being used at any given time. Workloads are variable usually.
It’s not that these datacenters can’t be loud, but rather that due to their lower power usage (compared to AI or Crypto before that) they don’t usually have to be.
The problem with AI datacenters is that they are designed to maximize their capacity and run full throttle 24/7 365. These facilities are bigger, are completely full to the brim with servers, and those servers are all working very hard nonstop. The thing about computers is, the stronger they are and the harder you push them, the more power they require to run. When electronic devices of any kind use more power, they generate more heat. Too much heat will also kill these electronic devices, so they need more cooling. Cooling apparatus makes noise, and the more you have of it, the louder it gets.
The TL;DR of it all is that AI datacenters are designed to maximize their compute capacity, which maximizes their power consumption, which maximizes the heat they generate, so you maximize the cooling, which maximizes the noise.
There’s a separate issue back the the “Consumes more power” step where some datacenters can’t be sated by the local power infrastructure so they have to find ways to supplement the power they get from the grid with additional noisy things like gas generators and such.
I do not like the recent use of the word “compute” as a noun, it’s such incredibly obnoxious boardroom bullshit speak. Compute is a verb. Otherwise great comment that explains the situation pretty well.
Over the internet? Cloud computing, like it always has, or just use the word computing, or processing power, or any of a dozen different phrases we already use for it. Every time someone uses the word compute as a noun a techbro jacks off onto their earnings report.
You can’t use the word “computing”, it’s a gerund that represents an ongoing activity, it can’t be unitized into finite units. “Processing power” is not a noun, it’s a phrase.
Compute is not a boardroom term, it’s an engineering term. It’s used by professionals because it’s efficient, practical, and makes the most sense linguistically in the context it’s used. It may be inelegant, but it’s a function term for an ugly, unnatural thing, it’s not supposed to sound pretty.
Traditional datacenters usually came in one of two general forms, Colocation datacenters where anyone can rent rack space in the building, and your hyperscaler datacenters like AWS and Azure.
Colocation Datacenters are usually only partially full of servers at any given time, and are often rented by private enterprises running small time web applications or even just internal tools.
The hyperscalers are worse, but the business model is overbuilding capacity but not all the compute is actually being used at any given time. Workloads are variable usually.
It’s not that these datacenters can’t be loud, but rather that due to their lower power usage (compared to AI or Crypto before that) they don’t usually have to be.
The problem with AI datacenters is that they are designed to maximize their capacity and run full throttle 24/7 365. These facilities are bigger, are completely full to the brim with servers, and those servers are all working very hard nonstop. The thing about computers is, the stronger they are and the harder you push them, the more power they require to run. When electronic devices of any kind use more power, they generate more heat. Too much heat will also kill these electronic devices, so they need more cooling. Cooling apparatus makes noise, and the more you have of it, the louder it gets.
The TL;DR of it all is that AI datacenters are designed to maximize their compute capacity, which maximizes their power consumption, which maximizes the heat they generate, so you maximize the cooling, which maximizes the noise.
There’s a separate issue back the the “Consumes more power” step where some datacenters can’t be sated by the local power infrastructure so they have to find ways to supplement the power they get from the grid with additional noisy things like gas generators and such.
I do not like the recent use of the word “compute” as a noun, it’s such incredibly obnoxious boardroom bullshit speak. Compute is a verb. Otherwise great comment that explains the situation pretty well.
Ok.
Which noun would you prefer for the overarching generalized concept of computation sold as a product?
Over the internet? Cloud computing, like it always has, or just use the word computing, or processing power, or any of a dozen different phrases we already use for it. Every time someone uses the word compute as a noun a techbro jacks off onto their earnings report.
You can’t use the word “computing”, it’s a gerund that represents an ongoing activity, it can’t be unitized into finite units. “Processing power” is not a noun, it’s a phrase.
Compute is not a boardroom term, it’s an engineering term. It’s used by professionals because it’s efficient, practical, and makes the most sense linguistically in the context it’s used. It may be inelegant, but it’s a function term for an ugly, unnatural thing, it’s not supposed to sound pretty.
It’s also not in the dictionary so it’s still bullshit, obnoxious, and honestly just sounds stupid.
It’s absolutely in the dictionary, it’s perfectly plain and inoffensive, and it sounds perfectly descriptive of what it is.