The decline of computers as a general-purpose technology

Fab numbers are astronomical. "TSMC recently announced plans for a $19.5B 3nm plant for 2022". How can the next computing paradigm even take off when all the oxygen is depleted in chasing "proven" technologies?

> We do mean that the economic cycle that has led to the usage of a common computing platform, underpinned by rapidly improving universal processors, is giving way to a fragmentary cycle, where economics push users toward divergent computing platforms driven by special purpose processors.

computing in my view is rapidly subliming away into the cloud. whether users will have much computer at all in front of them is questionable, or perhaps semi-vestigial. work/applications are headed into data centers. want photoshop today? even the old-fashioned install & run it yourself version has major cloud connectivity.

even within the world.of.consumer electronics, I don't see specialization as a trend. video game consoles have tended towards "exactly like a pc" over time. cell phones are pc's with the most proprietary/controlled drivers on the planet. 5g base stations are pc software-defined-networks plus gobs of general purpose dsps/sdrs. in the smaller computing/devuce world, arm and espressif and soon risc-v are cutting out larger & larger swarths.

thus far ai processors have had fairly wide ability to operate models in a cross-architecyure fashion. make sure you can run onnx or tf-lite. big only marginally customized gpus still have a huge presence. there is variety here but remarkably little end user differentiation.

where are the specialized systems this write up talks to? where are the fragmented systems? the door is opening as we try to re-learn how to make silicon foundries, how to do open source chip making, but I haven't seen what's being talked about here, specialization, bifurcation of capabilities.

but cloud? cloud is murdering the shit out of general purpose computing. applicationization of software turns software effectively hard to end users. we have no power, no ability to adapt or change or see the computing. it's 100% what is given to us. every system is 100% special/specialized, and in the post Pax Intertwingularis death of the api & interoperability, that means radical rigidity. systems are radically more specialized & in&general, but it's not a hardware problem like this paper asserts and it's not something on thebhorizon: it's already happened, it's already made the general.purpose computer obsolete, and it's at the software level, it's about where most of the world's computing is run, on special systems, on the clouds.

This is what I personally feel is just missing these days. I was never part of the leading edge of hobby computing.

I was very far off, geographically, growing up a teenager in India. But back then, 1998 - 2006 ish, I feel the Internet was a lot more about experiments. I would read up on some software or hardware group all the time. LUGs were way popular. I used to be on chat rooms from MIT or other unis every other day.

Nowadays most people I see of that age group are happy with a YouTube video on their smartphones. General purpose of computers is simply not visible. Not even to the level that I, a curious teenager, had experienced in those years from so far in India.

We have too many tall-walled gardens now.

Update: I should add I am happy with how powerful computers are these days, but they are not for the purpose of learning or tinkering the actual device or its software. It is just way to consume media for the mainstream.

Here you go - https://twitter.com/sndrv/status/1356884973979115522

This seems a little sensationalist to me. The GPU is a general-purpose device optimized for highly parallel workloads. One could just as easily say that a traditional CPU is a special-purpose device optimized for branch-heavy code at the expense of parallel throughput.

The current state of GPUs is based on Nvidia's decision to focus on GPGPU instead of being a DirectX accelerator forever. They specifically decided to make the device more general-purpose than it needed to be at the time. Presumably they thought there was a chance to start another virtuous cycle for parallel computing.

I don't think we are close to realizing the GPU's potential as a general-purpose device. Imagine the kind of software we will have when every phone has the equivalent of today's biggest desktop GPU inside it.

The next revolutionary general-purpose device will probably feel like a niche product at first, just as the 3Dfx Voodoo did in 1996.