www.anandtech.com
Huh, TSMC's being real careful with N2 it seems. Earlier I thought it was possible that they were slow walking the transistor density gains since trying both GAAFET and new power delivery simultaneously seemed a big enough task.
Introducing GAAFET with N2 starts up volume production in 2H 2025, so expect products in 2026...
Maybe they then try backside power delivery with the 2026 N2 variant, so product in 2027... if they're still that risk-averse, I probably should not expect significant transistor density gains either.
(meanwhile, Intel currently plans to attempt their own version of GAAFET and new power delivery with 20A in 2024)
Might as well spitball a bit what's bouncing around in my head about this sort of thing as it relates to gaming (more specifically focused on the big consoles and their succession):
(reminder, this is just rambling for fun from a rando outsider layman)
Remember those bullcrap rumors from TCL about Pro-style models for PS5/Series in 2023/2024 cause they wanted to drum up 8k TV sales? Of course that's not happening; I don't think it's even feasible for RDNA3 on N5 within the limitations of console design to deliver the upgrade expected from a 'Pro'. I don't think that there will be hardware matrix math acceleration in RDNA3, and I don't see sufficient raw grunt gains in a
economically viable monolithic design.
I think that at minimum, 'Pro'-level upgrade requires at least 2 nodes worth of transistor density/power savings and/or all-around improvement in reconstruction techniques. By all-around improvement, I'm referring to like how DLSS beats FSR 2 in image quality, reconstruction time, energy usage, and area usage. Anyway, timeline-wise, that is probably at least RDNA4 in 2024+. If RDNA4 does launch in 2024, I'm betting on MCM/high end only at start, with monolithic no earlier than 2025.
As for a 'PS6'/a 'full generation's worth of improvement as people have been trained to expect? N2's transistor density gains seem low enough that it'd probably take both N2
and hardware AI acceleration. And, thanks to AMD being in very dire condition prior to the success of Zen, they're way behind Nvidia in AI investment. Of course, the wildcard here is what Intel could potentially deliver in the back half of this decade. Or maybe Samsung bounces back in the GAAFET era, who knows. Maybe sheer transistor density + ok enough dedicated hardware could do the job.
As for Nintendo in a post-Drake world? Assuming Nintendo isn't trying some AR or VR system and opts for a 3rd major go at a Switch-type system?
shrug Wait and see what Nvidia's next killer feature is? Otherwise, depending on what node Drake is on, there might not be all that much room for raw grunt improvement in the near future.