Digitimes is confident that both fabs get part of the manufacturing cake, but TSMC gets the larger piece. Independently from Taiwanese sources, we can confirm that both TSMC and Samsung will manufacture the Ampere based solutions.
It was interesting to read that most reports expected Ampere solutions in Q3 2020, while everyone should know by now that Nvidia always introduces its next-generation HP AI chip at its GTC conference. The conference slipped for two months from March to May 15, and this is the time when Jensen will tell us all about this new massive and fast chip.
Samsung and TSMC manufacturing
A GPU gaming based solution is expected later, but this was always the case. Nvidia is doing well with RTX 2000 series, and between TI, Super and vanilla card is selling quite well, especially in the high-end market.
AMD is expected to jump in with a Ray tracing based solution later this year, but we are sure that Nvidia is not sleeping and has its next-generation RTX based products planned for the late part of this year.
The new name Hopper has been around for a while. The next-generation architecture is finally named after US citizen and scientist Grace Hopper. Grace was one of the first programmers of the Harvard Mark I computer.
Hopper is next
Back in 2017, Nvidia research posted a publication that the company is looking into MCM Multi-Chip-Module approach, and from where we stand, it is hard to know if these techniques will be applied to the Hopper.
Hopper is rumored to be a 5nm chip and comes in 2021, but this cadence might sound too optimistic. GPUs rarely jump from one manufacturing node such as 7nm to 5nm within a year, at least not in the same product category.
In any case, Nvidia has scheduled the GTC 2020 live online keynote for May 14 6.00 AM PST (9.00 AM ET, 15.00 CET), and Jensen will undoubtedly tell us more about the data center Ampere.