Meet Blackwell: NVIDIA's Powerful New AI Chip

The NVIDIA Blackwell GB200 chip, named after pioneering mathematician David Blackwell, is touted as the most powerful AI chip ever created, designed from the ground up to tackle the increasing complexities and computational demands of modern AI LLMS and applications. It's expected to ship later in 2024. 

"For three decades we’ve pursued accelerated computing, with the goal of enabling transformative breakthroughs like deep learning and AI," NVIDIA CEO Jensen Huang said at the developer conference. "Generative AI is the defining technology of our time. Blackwell is the engine to power this new industrial revolution. Working with the most dynamic companies in the world, we will realize the promise of AI for every industry."

The GB200 is actually two GPU dies connected by 10 TB/second chip-to-chip link. It's built on a custom 4-nanometer process and packs in 208 billion transistors. It has up to 20 petaflops of power. Compared to the current H100 chip, which can run at 4 petaflops, NVIDIA says Blackwell could reduce overall cost and energy consumption by up to 25x. 

You could even use the also-announced HGX B200 server board to link eight GB200 GPUs with networking speeds up to 400Gb/s. With multiple server boards and an NVLink switch, you could conceivably get up to 576 GPUs working together. 

Blackwell's capabilities could be pivotal for a wide range of applications, from deep learning and data analytics to autonomous vehicles and beyond, offering a glimpse into a future where AI is built into nearly everything. Huang told CNBC this week that each GB200 chip could cost $30,000 to $40,000, and the company had spent $10 billion developing it. 

All the big cloud-service providers, including AWS, Google Cloud, Microsoft Azure and Oracle Cloud are expected to offer Blackwell-powered instances, so companies shifting into more powerful AI won't necessarily have to run the hardware themselves. 

But it's not all about industrial uses. For consumers and computer enthusiasts, it's likely that some of this technology will find its way to consumer products, possibly even using the Blackwell architecture to power eventual next-gen 50-series NVIDIA RTX GPUs.