Tomorrow is gonna be nuts...Flurry of unconfirmed rumorsAll 14nm and below tools banned from China?All NAND/DRAM tools banned from China?All high end AI chips banned from China?High end DC CPUs like Genoa banned from China?All China fabless firms banned from EUV nodes?— Dylan Patel (@dylan522p) October 7, 2022
One of several articles on this:
https://www.reuters.com/article/usa-china-supercomputers-idCAKBN2R223O
Does the 14nm actually do something? Woulnt electron migration erode them faster thus you whoud need to run them at lower voltages and temperatures thus slower? Unless you vew them as disposable?
I’m talking shit, but yes they do do something… state of the art is 3nm/5nm
The -nm is just marketing nowadays, not tied to a physical characteristic, so the smallest feature size isn’t really 3nm or 5nm, etc anymore but each node shrink is still meaningful and “smaller than the last one”
And to your point, chips be going up to 1000 watts! (Invest in liquid cooling and short the planet.)
Interesting. I tougth it wass all marketing after all they had 8ghz celeron procesors back in the day. But isnt the limit to liquid cooling the interface between the chip and the fluid? It wold sem very hard to improve on that unles you got bigger chips?
Does the 14nm actually do something? Woulnt electron migration erode them faster thus you whoud need to run them at lower voltages and temperatures thus slower? Unless you vew them as disposable?
electron tunneling is no match for Moore’s law…
I’m talking shit, but yes they do do something… state of the art is 3nm/5nm
The -nm is just marketing nowadays, not tied to a physical characteristic, so the smallest feature size isn’t really 3nm or 5nm, etc anymore but each node shrink is still meaningful and “smaller than the last one”
And to your point, chips be going up to 1000 watts! (Invest in liquid cooling and short the planet.)
Interesting. I tougth it wass all marketing after all they had 8ghz celeron procesors back in the day. But isnt the limit to liquid cooling the interface between the chip and the fluid? It wold sem very hard to improve on that unles you got bigger chips?
Most gear is still air-cooled but that’s changing pretty quickly
Liquid cools like 3000x more efficiently than air, something like that
Also see ideas like this: https://www.ibm.com/blogs/research/2017/08/beat-heat-3d-chip-stacks-icecool/
I think IBM still has a 5Ghz mainframe chip out there but since multicore era frequency is mostly ~2-3 Ghz