Every time a major manufacturer makes a node transition, they have to decide which foundry partner to work with. These days, that comes down to one of two (possibly three) choices: TSMC, Samsung, and potentially Intel, who still apparently operates a client foundry business but hasn’t made any major customer announcements in quite some time. For the most part, then, this is a two-way race between TSMC and Samsung.
TSMC has done well on 7nm, locking up a lot of the major business and early customers. The two firms pursued different strategies on 7nm, with Samsung opting to wait to introduce the node until its Extreme Ultraviolet Lithography (EUV) was ready, while TSMC introduced 7nm as a conventional lithography node with plans to roll out EUV for mass manufacturing next year.
Samsung has been a little light on customers, though we know IBM is going to do some building with the foundry long-term. During a press conference in Seoul this week, Nvidia Korea chief Yoo Eung-joon confirmed that the two companies would collaborate on future GPU designs.
“It is meaningful that Samsung Electronics’ 7-nanometer process would be used in manufacturing our next-generation GPU,” said Yoo. “Until recently, Samsung has been working really hard to secure (partners) for foundry cooperation.”
Nvidia’s 12nm Turing architecture, built at TSMC.
The exact terms of the deal were not disclosed, nor exactly which parts will be built at Samsung, but it’s possible that Nvidia will go all-in with its new partner. Designs for one foundry can’t be easily or simply ported to another, which means building the same chips at two leading-edge foundries means doubling up on some design work. Nvidia might opt to tap both companies, but using just Samsung would simplify some aspects of the design process. Samsung’s 7nm pricing is reportedly extremely competitive, in the hopes of goosing its design wins.
As for when we might see 7nm chips from Nvidia, that’s anyone’s guess. AMD is launching its 7nm Navi chips this coming Sunday, and Nvidia has already answered them by tweaking prices on its existing RTX cards. But we also know that Navi is going to make the leap to “Big Navi” next year, and that’s one place where Nvidia will likely want to have a 7nm response of its own ready. If AMD’s 7nm technology proves potent, Nvidia won’t want to meet it with its own aging Turing family — and if AMD is still lagging in terms of overall competitive standing, Nvidia will be able to drive that home with its own new launch.
Either way, we don’t expect to see Turing hold the market as long as Pascal did. Pascal holds the record for the longest-lived GPU architecture in the consumer space, having dominated from May 2016 through September 2018. We have absolutely no information on when Turing’s successor might launch, and there are definitely lead times when bringing new processes to market, but 12-14 months from now until launch is a pretty reasonable bet.