Is Moore's law dead?

[BurakZG]

I've heard this (incorrect) claim many times.

Watch this interview with Christophe Fouquet, CEO of ASML, talking about the "More's law" and other topics.
https://www.youtube.com/watch?v=hvQvrLjlJN8&list=PLzYp5zh-CrBUX_4WnPCO1jj5v6jMuQjKe&index=2

Hint: If you don't know what's ASML, maybe don't talk about Moore's law ;-)

[A&A]

RTX Titan (2000 series) and RTX5090, both have similarly sized chips. The age gap between both are 6 years and the trasistor count difference is around 5 times.
It is somewhat alive.

[shiel]

Buy into that BS all you want. The fact the Nvidia gave such a massive increase between the 3090 and 4090 shows what BS that is. The 5080 looked lackluster on paper and lived up to that.

[Andrius227]

Originally posted by shiel:

Buy into that BS all you want. The fact the Nvidia gave such a massive increase between the 3090 and 4090 shows what BS that is. The 5080 looked lackluster on paper and lived up to that.

5080/5090 are still much more advanced than 40 series, just not in the way most people would have preferred.

They shrunk the pcb’s significantly which is impressive although most people don’t care.

[C1REX]

Originally posted by shiel:

Buy into that BS all you want. The fact the Nvidia gave such a massive increase between the 3090 and 4090 shows what BS that is. The 5080 looked lackluster on paper and lived up to that.

3090 to 4090 uplift was exceptional because Nvidia moved from crappy Samsung 8nm to amazing TSMC 5nm. TSMC is already the best in the world so they couldn’t make such 2generations leap again.

[AmaiAmai]

Don't forget a lot of it is also software. As time goes on, we move away from certain trends and move towards others.

That can also skew data because how software is written might favour one change in architecture over another, and in some cases when instructions are removed or disabled (for various reasons including security flaws) it can lead to negative performance or a smaller uplift than expected.

This sometimes happens right before a release of a hardware and isn't reported and leads to less performance than expected v. the claimed performance. Most flaws in chips that are hardware-level like that are not disclosed for 12 months or more after launch because it takes time to assess whether there are more.

[His Majesty's strawberry cart]

putting more rivets into a steam boiler is only gonna make it a server