Attempts to challenge Nvidia's hegemony in the smart chip market

The second-generation, domestically produced smart chip from social media giant MITA will be available for use by the end of this year.

The Meta project is part of a bigger and more comprehensive plan that also entails developing an artificial intelligence infrastructure for the business to guarantee its use in the services it offers on WhatsApp, Facebook, and Instagram.

The first chip model was unveiled by the business in May 2023, and production will start the following year. Previously, the company used NVIDIA H100 processors, which are specialized chips to speed up artificial intelligence training and inference processes, to control the operations of its gadgets.

Reuters reports that the corporation plans to purchase 350,000 of these chips for its artificial intelligence applications, bringing its overall chip count to 600,000.
Mark Zuckerberg, the company's founder, wants to move away from NVIDIA products and toward locally produced goods. Additionally, he wants to end NVIDIA's stronghold on the artificial intelligence chipset market.

Meta chipset of the second generation

Utilizing a technology known as 5nm architecture, the Meta chipset of generation II is 3.5 times more advanced than its predecessor and has a network of 8 rows and 8 columns of processing components.

.

Meta is also working to support the chip infrastructure by developing “racks” that can accommodate 72 accelerators simultaneously within the “server.” Modern accelerators can operate at a frequency of 1.35 GHz, compared to 800 MHz in the old ones.

The modern design aims to provide more computing power, bandwidth, and memory capacity to the chips, and as such, the pace of training artificial intelligence models will increase to unprecedented levels.

Modern servers have a higher chip capacity, which increases the capacity of data processing and training operations in artificial intelligence (meta).

Breaking Nvidia's dominance in the world of chips

Not only is Meta aspire to break Nvidia's dominance in designing and building artificial intelligence chips, but even Intel, which is famous for manufacturing old traditional chips, which lagged behind in the race in manufacturing the basic requirements for artificial intelligence chips, aims to return strongly after announcing the "Gaudi" chips. (Gaudi) in a press release issued earlier this week.

Intel claims that its AI chip can train AI models 3 times faster than the popular NVIDIA H-100 processors. Gaudi's chips also use a 5nm architecture and consist of two integrated main processors.

While the Tensor processing units of the technology giant Google are the only competitor to NVIDIA, although Google does not aim to sell its smart chips, but rather offers them to developers through the Google Cloud platform.

Source : Reuters + websites