Hacker News Reader

Comment by 1letterunixname | Hacker News Reader

Expecting hockey stick growth indefinitely is patently ridiculous.

Surely there will be an inevitable insurmountable plateau reducing cost and performance gains to smaller improvements.

At some point, different technologies will have to be seriously considered such as photonics.

There is an argument that it has to hit the plateau first before a new technology can take over, the problem being that if the current tech is really scaling by 2x every 2 years, a new technology that is 10 years from commercialization has to be more than 30x better than current technology to be worth considering. However if current scaling slows down, the extra delta required is reduced.

I seem to remember this was an argument made with Josephson Junctions back in the day. They were potentially better, but always far enough out that traditional technologies would exceed the projected performance by the time they launched.

So maybe we need to hope for current trends to plateau if we want something new to replace it.

Sure, but I've heard people expect it to plateau soon for 40+ years now.

It will be true some day, and maybe that's now, but I've stopped thinking we'll know before it happens.

Aren't there still massive gains on the table? I'm totally naive on this but I recall hearing about other materials, such as graphite iirc, that could achieve 100x CPU hz.

According to Jim Keller, yes, there is. If my memory serves, he said so in a recent Lex Friedman interview.

Edit: here is the clip https://youtu.be/c01BlUDIlK4

Thanks, this is pretty much what I was getting at. While "Moore's Law" might strictly mean transistor density, "performance" is about many things.

We're already limited by lightspeed propagation, so it's hard to see how this could be the case.

I assume what you mean is that the CPU may get faster but would ultimately stall waiting for data, which can only move so fast. Presumably to continue feeding data to the CPU we'd want to be much more aggressive about caching and prefetching.

Another option would be to not go 100x faster but to go 10x faster and then take the improved cooling + energy usage as a win.

Does the article say anything about indefinite hockey stick growth? It seems a lot more measured, just saying that Moore’s law will keep going for at least another decade or so.

Reply on news.ycombinator.com