It will be in a big event as expected and not just any event. The presentation of Hopper as a revolutionary architecture for NVIDIA already has a date set by the company and is closer than we thought, which makes us suspect that those of Huang could also advance Ada Lovelace and launch a double attack against AMD. When will we have all the details of Hopper?
We have debated at length about the leaks that have been coming out where one curiously questioned what NVIDIA could bring to market. Apparently he was wrong, since he questioned that there were two different architectures between GPUs for AI and DL compared to desktops and laptops. Now it seems that this is discarded in the wake of the presentation.
NVIDIA Hopper at 5nm: we will see the details at the GTC
The greens have hit the key again, leaving CES 2022 for more immediate products (the RTX 3090 Ti is obviously apart) and they are going to leave the main courses for significant events and more in line with their philosophy.
It will be at the GTC 2022 where the company shows all the details of Hopper and at the same time possibly some more innovation or decoy of what is to come. The presentation is fixed between the March 21 and 24 of the same monthso we really are just over 30 days away from all the architecture stuff being revealed to us.
It will once again be an online event due to the sustained concern that NVIDIA has shown with COVID-19, where the highlight of the event is expected to be the day 22 as Huang, CEO of the company, will debut with a speech. The rumors about whether said speech will be again in the virtual kitchen that we have been seeing have already been triggered, because what the greens are going to present is such a qualitative leap as was Ray Tracing and DL with RT Cores and Tensor Cores , as well as the RTX as a range.
NVIDIA GH100 and its lithographic process
The rumors are having their effect and based on them it seems that we have a vague idea of what exactly the conference will focus on. The new GH100 chip It will logically be the culmination of the Hopper architecture and the epicenter of the products presented, but for the first time at an NVIDIA event it seems that a company and partner will be emphasized in the process.
Logically this is TSMC, since its 5nm process that will accompany the architecture and Ada Lovelace is the key piece that will allow NVIDIA to prevail over Intel and the news that they will present should be ahead of AMD’s MI.
What is expected is that the GH100 chip will be the largest chip ever created in terms of dimensions with just over 900mm2 and SM doubling is also expected with up to 288 in total, mostly thanks to packaging LCM, the main novelty of this architecture as such. Ada Lovelace, on the other hand, will be a monodie chip and not based on chiplets, where, on the other hand, it is expected that it will be the last one that NVIDIA launches with this technology in order to be able to face AMD and Intel already in 2024.