Slashdot Mirror
How Many Computers Does the World Need? (ft.com)
An anonymous reader shares a report: It is almost a decade since Rick Rashid, then head of research at Microsoft, posed that question and ventured his own answer: no more than a few, at least to handle the vast majority of the planet's digital workload. Back then, he thought, it was possible to discern the emergence of a small group of companies that would run those computers. The give-away was that a fifth of all the servers sold in the world were already being purchased by a clutch of US tech groups that included Amazon, Google and Microsoft.
[...] Of course, "how many" is a trick question when it comes to the distributed computing systems being built by today's tech giants. There are many nodes to these octopus-like systems, each with its own silicon brain and information-processing capabilities. But they are connected to a greater whole. One sign of just how far their tentacles are starting to reach came this week with Google's announcement that it has designed an AI chip to run in smartphones and other devices. Google's TPUs -- processors that are optimised to both train deep-learning algorithms and then apply them to make inferences from new data -- are already a key part of its data centre infrastructure. The new low-power version of the TPU can make inferences in "edge" devices, far from the computing core, and will be an important element in making sense of the world's data.
[...] The rise of the global computers raises many questions, but two stand out: will they comprise a truly competitive market, or come to look like the more Balkanised "platform" markets in the consumer world? And what will it mean for so much computing power to be concentrated in a handful of private companies? The good news is that the cloud landscape is shaping up to be a competitive one, at least if competition can be said to truly exist between oligarchs.
[...] Of course, "how many" is a trick question when it comes to the distributed computing systems being built by today's tech giants. There are many nodes to these octopus-like systems, each with its own silicon brain and information-processing capabilities. But they are connected to a greater whole. One sign of just how far their tentacles are starting to reach came this week with Google's announcement that it has designed an AI chip to run in smartphones and other devices. Google's TPUs -- processors that are optimised to both train deep-learning algorithms and then apply them to make inferences from new data -- are already a key part of its data centre infrastructure. The new low-power version of the TPU can make inferences in "edge" devices, far from the computing core, and will be an important element in making sense of the world's data.
[...] The rise of the global computers raises many questions, but two stand out: will they comprise a truly competitive market, or come to look like the more Balkanised "platform" markets in the consumer world? And what will it mean for so much computing power to be concentrated in a handful of private companies? The good news is that the cloud landscape is shaping up to be a competitive one, at least if competition can be said to truly exist between oligarchs.
Three computers for Elven-kings under the sky.
Seven for the Dwarf-lords in their halls of stone.
Nine for mortal men doomed to die.
One for the Dark Lord on his dark throne
In the Land of Mordor where the shadows lie.
One computer to rule them all, one computer to find them
One computer to bring them all, and in the darkness bind them.
In the Land of Mordor where the shadows lie.
Article is paywalled.
Also, define "need." We could go back to the client (dumb terminal)/server centralized model of computing, but even the "dumb terminals" are computers in this day and age.
And not everyone wants to hand their private data over to a megalith like Microsoft, Google, or Amazon, so there's a place for computing devices that don't HAVE to act as terminals.
I mean, you could always take Uber, be recorded, tracked, and advertised to. But sometimes, you just need to drive that red Barchetta on a winding mountain road.
Also, the summary implies the article is stupid.
If, and I mean if, this is the sort of question to ask, it would ask how much computing does the world need. Computers run the gamut of speed from slow to fast, and capabilities from microcontroller to high-end server.
Not to mention the capabilities of a display adapter used for rendering or as a general-purpose parallel computing device.
From the viewpoint of information theory, all computers are equivalent in the sense that they can be shown equivalent to a universal turing machine, so the question isn't even definable in the mathematical sense, but we could assign an arbitrary measure and time scale to make it meaningful to humans.
For example, "millions of 8-bit additions per second" sounds like a reasonable low-level measurement (compare to "mm", for instance).
Then one could ask "how much computing does the world need".
And now we need to define "need". Just about every electronic device you can purchase today has an embedded microcontroller with a fixed program. Clocks and watches have little computers inside them.
All cars need computers to manage their inner workings, and most of the world doesn't own a car but would like to. The average car has about 30 computers.
Has that been included in the calculations?
Computer time used to be metered. To take a course in college you were allocated a (generous for the application) number of CPU minutes to do your homework (both computer and non-computer classes). To do a study you were allocated a number of CPU minutes to use for the calculations.
Today, compute time is so cheap we don't to meter it - we meter the amount of electricity that is used, or the annoyance of keeping the hardware running.
How much computing would people use if they had access to an unlimited supply?
We don't really know, because we're still on the leading edge of the bell curve. We yet to saturate even one person's use of computing - we still don't have ubiquitous AI in self-driving cars and factories.
Article is stupid. It's impossible to answer the question today, and they're even asking the wrong questions.