New information arrives on NVIDIA Hopper, the future GPU architecture that will succeed the current Ampere. From the looks of it, it would have lagged within NVIDIA’s roadmap. Let’s see what this news might imply for the graphics cardboard.
NVIDIA Hopper would have been delayed: will they have to see the Ampere problems?
The information comes from the well-known leaker kopite7kimi, so it is nothing more than a rumor, but we must bear in mind that this mysterious user has a long history of successes in terms of leaks related to NVIDIA, so of course, he has his sources.
I'm afraid the Next-Gen GPU (Hopper) based on MCM was delayed. Is Jensen going to draw a new roadmap instead? It looks very probable.
— kopite7kimi (@kopite7kimi) December 10, 2020
The leaker made the claim in a tweet, also hinting that the NVIDIA roadmap could be revamped .
Based on what is known so far, Hopper will have as its main feature to use an MCM design. This is, “ Multi-Chip Module”, which basically means that the same chip is divided into several interconnected modules, in the same way as in AMD Zen 2 CPUs, where we have the so-called “chiplets”.
The possible launch date of Hopper after the delay is not clearly known, as they were expected in 2021, but we do know his MCM design and that he will use the 5nm process of TSMC.
What we do not know are the reasons why it would have been delayed: could it have to do with the problems of Ampere, or was it something else entirely?
This delay could make Intel the first to introduce a GPU of this type, something they already have in mind to bring a great beast to the market, and could even lead AMD to get ahead of NVIDIA in it, as there are already rumors that the Future RDNA 3 could employ a chiplet design.
As always, NVIDIA architectures are named after an important historical figure in computing. In this case, that of the scientist and military Grace Hopper, a pioneer in computer programming, and whose work has been essential for all the technology we know today.
Well, if this information is confirmed and Hopper will take longer to leave, will NVIDIA bring us something special in the meantime? Will we have some kind of “Ampere refresh” or other intermediate architecture?