Comment by immibis
3 months ago
Okay? There's a lot of chips you can make that aren't the cutting edge. You don't need a 4090 to do AI, as evidenced by all the AI we did before the 4090. You definitely don't need a (random Intel chip) 14900HX to do general-purpose computing, as evidenced by all the general-purpose computing we did before the 14900HX.
For that matter, the 14900hx was already based on a refined 7nm production process, which China already has started using, though maybe not as effectively yet. As you mention, prior to the 4090's 3090 was on an 8nm node, already behind current China capabilities.
If each node provides a 10-15% improvement in power, performance and area, how many of those need to compound until your already uncompetitive 7 nm is 10x less efficient, slower and more expensive?
Being behind doesn't mean they're permanently stuck where they are today - but aren't our processes running into the wall of soon trying to make transistors smaller than an atom?
> Being behind doesn't mean they're permanently stuck where they are today
Without EUV, they very much are.
> but aren't our processes running into the wall of soon trying to make transistors smaller than an atom?
No, the finest pitches are still in the low double digit nanometers in 2 nm processes. The "2 nm" nomenclature hasn't denoted a physical dimension for decades.
2 replies →