Although it seems that things are already returning to normal when it comes to the COVID-19 pandemic, when it comes to technology fairs the presentations continue to trend towards the digital / virtual. The fair GTC 2022 is already here (starts today and ends on the 24th), and NVIDIA You have already confirmed your attendance at it with a key note which will take place tomorrow the 22nd, where, as we are going to explain to you below, it seems more than confirmed that they will present their Hopper architecture.
Hardware fans have been waiting for quite some time for NVIDIA to make a move on its next generation of GPU architecture, called Hopper, and from which we have not stopped receiving leaks in recent weeks and months. It seems that the GTC 2022 will finally be the stage in which the company will put an end to the wait, or so they have dropped it in an announcement on their blog under the title (own translation) «Hopped up: the CEO of NVIDIA and AI leaders will talk about the next generation at the GTC.” Obviously the expression “hopped up» (instead of Hooked up, which would mean «connected») refers to Hopper.
NVIDIA Hopper at the GTC, what do we expect from it?
One of the great novelties that -in theory- the NVIDIA Hopper architecture will bring to the market is that it will be the first time that the company incorporates MCM (multi-chip module) design with up to two chiplets. In addition, according to the rumors that have been arriving in recent months, we would also be talking about the largest GPUs in the company’s history, with a size greater than 853 mm², which if the rumors are true would mean that the flagship GPU of the generation (GH100) would have 2.6 times more transistors than Ampere, reaching 140,000 million.
The complete configuration of the GPU chiplet could then offer up to 144 Stream Processors and up to 18,432 CUDA cores, so if each graphics card incorporates two of these chiplets, it gives us no more and no less than 288 Stream Processors and 36,864 CUDA cores. . Logically this would be the architecture of the next generation GeForce RTX 40 Seriesand as such we will see “casted” chips to adapt to the different models that will be launched on the market, as well as to meet the performance and consumption requirements (remember that its predecessor was already casted and had a 16% of the chip disabledplus one of the memory chips was missing).
In addition, past rumors also told us that the GH100 chip was not the only one that the company was working on, since the GH202 chip has also been talked about, although it is true that for now absolutely no details about this chip are known. in particular, so it could be anything (this name came from a supposed hack to the firm’s servers).
NVIDIA will make an appearance during GTC 2022 tomorrow, March 22, 2022 at 16:00 Spain time (it is 8:00 am PST time).