To cement its supremacy over the AI sector, Nvidia has unveiled its flagship AI chip, the B200, the most powerful single-chip Graphic Processing Unit (GPU) with 208 billion transistors.

The B200 is part of the new Blackwell architecture, which includes six revolutionary technologies that enable AI training and real-time large language model (LLM) inference for models up to 10 trillion parameters. The B200 chip is designed to reduce AI inference operating costs and energy consumption.
Nvidia also launched the GB200, which combines two B200 GPUs with a single Grace central processing unit (CPU). The GB200 can deliver a 30x performance increase for LLM inference applications compared to the same number of Nvidia H100 GPUs and decrease cost and energy consumption by up to 25x.

Why does it matter?

The B200 chip and GB200 super chip are expected to be adopted by several major tech firms, including Amazon Web Services (AWS), Google, Meta, Microsoft, OpenAI, xAI, Oracle, and Tesla. Nvidia claims that its new Blackwell chips can deliver a 30x performance improvement when running generative AI services based on LLMs such as OpenAI's GPT-4 compared with previous GPU architectures, all while using 25x less energy. The new Nvidia's Blackwell offering is expected to power the training of multi-trillion-parameter AI models that will make today's generative AI products obsolete, fueling the AI race even further.