Nvidia unveils the world’s most powerful GPU (AI-focused)
(San Jose, USA) Nvidia announced on Monday (18/03) the Blackwell platform, developed for those who need great processing power in artificial intelligence tasks with more than 1 trillion parameters. The company’s founder and CEO, Jensen Huang, ranked it as the most powerful GPU ever. However, it was not designed for graphics in games, but for activities of the so-called new industrial revolution.
The base GPU is the Blackwell B200, with 208 billion transistors, a huge amount. That’s 20 petaflops of processing power. It succeeds H100, famous for having Mark Zuckerberg among its buyers (with hundreds of thousands of units on the way to Meta).
It’s twice as fast as the Hopper architecture, according to the executive, which was launched two years ago. I attended the launch at the SAP Center and could see the astonishment of the audience when the screen showed the video of a datacenter forming with the technologies of the Blackwell platform.
Nvidia believes that the Blackwell architecture will have an impact on engineering simulation tasks, electronics design automation, data processing, drug development, quantum computing, and generative artificial intelligence. It should be employed in Amazon Web projects
Services, Dell Technologies, Google, Meta, Microsoft, OpenAI, Oracle, Tesla, and xAI.
In fact, throughout the GTC 2024 congress, executives have positioned Nvidia as the only AI company that talks to everyone else – from language model developers to suppliers of this technology to other businesses.
Nvidia also announced the GB200 superchip. It involves two Nvidia B200 and an Nvidia Grace CPU, connected at the rate of 900 GB/s via an NVLink (with low power consumption).
Jensen Huang’s company did not elaborate on pricing or availability. As GTC is a conference mostly for developers, it is unlikely that they will announce GPUs for the gaming market.
