Nvidia signaled the imminent launch of its Ampere GPU architecture at GTC 2020, the next-generation datacenter processor.
NVIDIA is preparing to announce its Ampere GPUs following CEO Jensen Huang's recent tease at the company's quarterly earnings conference for GTC 2020 in San Jose, where he stated "you won't be disappointed." Previous leaks have indicated a March 2020 timeframe for launch. During the earnings call, Huang reported "excellent" quarterly results with strong demand for NVIDIA RTX graphics and AI platforms, record data center revenue, and noted that NVIDIA RTX is "reinventing computer graphics" while driving upgrade cycles in gaming and professional graphics. He invited all attendees to GTC, where he promised additional announcements about key industry trends.
The shift to Ampere represents a significant evolution from NVIDIA's current Turing architecture. The company has noted that gamers are increasingly shifting to its Turing GPUs as prices normalize, though NVIDIA has not yet reached the heights of its cryptocurrency boom era. Currently, the Turing GPU is capable of raytracing at 1080p 30 fps for light to moderate path ray tracing workloads, with Ampere expected to advance these capabilities further. NVIDIA has maintained the 16nm process (12NFF node, which is a rebadged 16nm FF node) for an extended period, but with AMD already on 7nm, competitive pressure is intensifying.
The Ampere GPUs will be based on Samsung's 7nm EUV process, delivering both performance and power efficiency advantages. The 7nm EUV approach is considered easier to fabricate than standard UV multi-patterning efforts, functioning as a reset of the difficulty curve with the new light source, though it will require extensive retooling. At minimum, the process transition is expected to yield a 50% increase in performance all things considered on a watt-for-watt basis.
NVIDIA's decision to shift from its traditional foundry partner TSMC to Samsung represents a significant strategic move. As one of TSMC's biggest and most loyal customers since the dawn of modern gaming technology, NVIDIA's departure signals either a superior financial offer from Samsung or confidence in Samsung's technical advantages. TSMC's current 7nm process is not based on EUV, though they have an EUV node planned, whereas Samsung's process is already EUV-based. Additionally, NVIDIA may face capacity constraints at TSMC, which prioritizes customers like Apple and is heavily utilized by AMD.
Samsung stands as the logical partner for NVIDIA's ambitions, possessing ample foundry capacity and the financial resources to address yield challenges. With Intel's foundry capabilities under question and TSMC heavily constrained, Samsung remains the only viable leading-edge foundry alternative available. GlobalFoundries exited the leading-edge race earlier in the year and would not have been under consideration regardless.