General Discussions

[black_zion]

This is interesting. Even if they don't use it, it keeps AMD from using it, which could force RNDA3 to be 7nm or face abysmal supply at launch. Like the article says, nVidia has a much larger pile of cash than AMD so they can do it. It also means that we could be seeing Ampere's successor in 2021 which could mean AMD has to rush out RDNA3 or face falling even more behind, and AMD is already having to madly play catchup.

https://wccftech.com/nvidias-hopper-architecture-will-be-made-on-tsmcs-5nm-process-launching-2021/

[pokester]

These are the caveats of having so few FAB choices these days. I wonder if AMD wishes they had never spun off their fabs? Of course along with selling their building I doubt it was much of a choice and more of a necessity. 

[black_zion]

There are perks to being a fabless company though, namely that you don't have to spend the money in R&D to develop new processes, don't have the overhead to own and run a fab, aren't tied to a single fab so you can shop around for the best deal, and aren't responsible if a power outage costs thousands of wafers.

With the current political climate though, I think it would be a good idea if AMD and Samsung partnered up with a fab in the USA. Samsung could trade their mobile SoC division to AMD with the stipulation that they are always at least as efficient and fast as Qualcomm and Apple. In exchange AMD fronts the money for an equity stake, and gets an exclusive contact with Samsung to fab their chips at a rate which beats TSMC by a certain margin. AMD is sitting on a billion dollars in cash, chump change compared to Intel and nVidia but that would be enough for a decent amount of equity share in a fab. Texas Instruments spent $3.1 billion on a new fab in Texas last year for 300mm wafers in an 837,000ft² site, so that could be a deal they may look into. They would also no doubt be able to steal some orders from TSMC. It would be a profitable venture.