According to a report from DigiTimes, Nvidia will fab its upcoming Ampere architecture, which is expected to succeed Turing, on Samsung's 7nm EUV process rather than on the 7nm process from TSMC, which has been Nvidia's foundry partner for years. Ampere is expected to launch in 2020, though how the architecture differs from Turing isn't clear yet.
It was originally expected that Nvidia would be using the 7nm process from TSMC, which has found its way onto Apple iPhones and AMD CPUs and GPUs, but according to EETimes, Samsung has apparently "aggressively undercut" TSMC. That means the company could convince Nvidia to fab Ampere on Samsung's own process rather than the process of its foremost competitor. Samsung gaining Nvidia at TSMC's expense is a big power play for the company.
Earlier this week, Samsung also made a deal with AMD for its upcoming RDNA architecture, where Samsung gained access to AMD's IP and can now fab GPUs with AMD technology. Though this deal differs in that Samsung is making these chips for their own devices and not for AMD, it does show how keen Samsung is on leveraging its foundries and 7nm process in order to make partnerships with companies such as AMD and Nvidia, and even, at least in the case of Nvidia, take away some of TSMC's partners.
Production capacity might also play a part in Nvidia's choice to switch to Samsung. TSMC's 7nm node is seeing a very high amount of demand, especially from Apple and AMD, which recently launched its Ryzen desktop CPUs and EYPC server CPUs. By using Samsung's 7nm EUV process instead of TSMC's 7nm process, Nvidia might be able to have more supply since it can take years to adjust to high demand and build new foundries and facilities, as can be seen in the case of Intel's shortage.
