Nvidia drops a real bomb. “New Industrial Revolution”

Nvidia is one of the “hottest” technology companies in the world. The company from San Clara has about 80 percent of the AI ​​systems market share, and there are queues of people waiting for its most popular H100 and A100 chips. Today, Nvidia’s solutions power models of generative artificial intelligence such as the famous ChatGPT (OpenAI), Copilot (Microsoft) or Gemini (Google). Meta, the owner of Facebook, intends to buy as many as 350,000 A100 chips this year.

Experts point out that Nvidia is winning the war for the artificial intelligence market today, because its largest competitor AMD focuses almost exclusively on the consumer market. And although companies such as Google and Microsoft are thinking about producing their own systems to become independent from Nvidia, currently the Santa Clara company is the only one that has the know-how and technology to produce AI chips on a mass scale.

Hopper is dead, long live Blackwell

Two years after the premiere of the Hopper architecture, Nvidia decided to refresh a series of graphics chips for generative artificial intelligence models. Although the word “refresh” is an understatement here. We are talking about a gigantic increase in efficiency.

On Monday, the Greens presented a new series of AI chips based on the Blackwell architecture. The company boasts that they are seven to thirty times faster than H100 systems, and at the same time up to twenty-five times more energy efficient.

Blackwell GPUs are the engine powering the new industrial revolution

– said Nvidia CEO Jensen Huang during the GTC conference in San Jose. “Generative AI is the defining technology of our times. By collaborating with the world’s most dynamic companies, we will realize the promise of bringing AI to every industry,” added Huang,

NVIDIA GB200 Grace Blackwell Superchip NVIDIA

Nvidia is preparing five different Blackwell systems for its customers. The company boasts a performance jump in AI calculations from 4 petaflops to 20 petaflops. This additional computing power is intended to enable companies in the industry to train larger and more complex generative artificial intelligence models.

The first chip to be made available this year is the GB200 Grace Blackwell Superchip. It consists of two interconnected B200 graphics chips and one central processor based on ARM architecture.

Nvidia also announced that it will offer its partners entire GB200 NVlink 2 servers consisting of 72 Blackwell GPUs and other components. These will be ready-made solutions enabling the training of AI models.

One of Nvidia’s partners, Amazon, has already announced that it will build a server cluster based on 20 thousand (sic!) GB200 chips. New chips will also be available, among others: to Google, Microsoft, OpenAi, Oracle and even Elon Musk’s Tesla.

Currently, there are no better AI chips on the market than those produced by Nvidia

– Elon Musk said,

NVIDIA GB200 Grace Blackwell Superchip Nvidia

Blackwell offers a huge leap in performance and will accelerate our ability to deliver cutting-edge models

– noted Sam Altman, head of OpenAI and one of the creators of the ChatGPT service.

Traditionally, Nvidia has not revealed how much the Blackwell systems will cost. As it turns out

Despite the prohibitive prices, Nvidia has no problems selling its solutions, for which there are still queues of people willing to buy them. Last year, the waiting time for units was up to 11 months.

The premiere of Blackwell chips did not have a significant impact on Nvidia’s share prices on the stock exchange. On Monday, in after-hours trading, the company’s shares increased by 0.70 percent, which shows that this information was previously discounted by the market. The company’s market capitalization is currently $2.18 trillion, making it the third most valuable publicly traded American company – behind Microsoft and Apple.