Hacker News Reader

Hacker News

Comment by niemandhier | Hacker News Reader

Full thread

niemandhier

Triple product (efficiency ) has increased faster than moors law for the last 50 years.

Still people make jokes about fusion research, some things just take time.

I recommend this excellent review of the even more excellent book „The future of fusion energy“

https://www.astralcodexten.com/p/your-book-review-the-future...

View on HN

> Triple product (efficiency ) has increased faster than moors law for the last 50 years

Fusion research progress is underappreciated. But Moore's Law is for an existing industry. Prior to that, it took 10 ^ 6+ improvements in various technologies to make computing possible.

Not sure what you mean by that. There was a bunch of computing technology long before photolithography or even transistors. And mores law was coined in 1965 when individual chips had far less than 10^6 transistors while the first one was made by hand.

Indeed, I often think if I were transported to the past I would use rivers and dams to make logic gates.

Similarly, I would like to think that if I had been born a hundred years earlier, I would have invented something like MONIAC.

https://en.wikipedia.org/wiki/Phillips_Machine

If you were transported to the past with the knowledge you currently possess as a result of the work already done, sure.

Developing it from scratch is a whole other matter.

Moore's law is self propagating: improvements in compute beget improvements in compute by improving the computers used to design compute devices. fusion, while in an impressive bootstrapping phase, does not get that acceleration until commercial break-even.

Untrue on both counts. Moore's law is not due to improved computing power in the design process, it's due to an incredibly long series of process improvements which become easier to develop because of the increased understanding of the problems they are seeking to solve.

Likewise advancements in fusion beget advancements in fusion by increasing understanding of the challenges faced.

> improvements which become easier to develop because of the increased understanding of the problems they are seeking to solve.

Surely this is aided by ever more sophisticated computational models? Maybe not the dominant factor in Moore's law, but maybe not negligible?

Probably not. Materials science research is much more dependent on analog electronics, specifically measuring voltage signals, than on digital electronics.

Modern computers are certainly a nice quality of life improvement for researchers though.

That said, commercial production lines might be a different story. Now I'm wondering what the minimum viable computer system would be to implement the firmware for ASMLs machines, for example. Probably more than vacuum tubes but I'm guessing you could pull it off with an old 8 bit chip. That's just a guess though.

Yeah, it's aided by better computational models, but not appreciably by having those models run on chips with smaller feature sizes.

If a time traveler from the future gave everyone at ASML and TSMC computers that ran 4 times as quickly so they could run their models faster, but no one could take a look at how they worked or were made, it wouldn't have any noticeable effect.

Moore's law is empirical, and do you honestly think an asml duv machine could exist with 1980s computing power?

> Moore's law is empirical

So?

> and do you honestly think an asml duv machine could exist with 1980s computing power?

I think something comparable could be, yes. The difficulty is in the precision, not the controls. Hell, the controllers in the actual machine likely only have 90s level computer power.

moore's law is just the observation that chips are 2d: linear shrinks produce exponential benefit.

I can't see anything like a linear/square relation in fusion reactor design (even accepting that ML's premise of shrinks being linear in time is not a law, just something that sometimes happened, and sometimes didn't...)

Reply on news.ycombinator.com