Jensen Huang, founder and CEO of NVIDIA, explained the technological advancements that will be brought about by increased computing performance, including the new Blackwell computing platform, in his GTC keynote speech held at the SAP Center in Silicon Valley on the 19th.

About 11,000 people attended the event that day, and tens of thousands more watched the keynote speech live online.

“Accelerated computing has reached an inflection point, and general-purpose computing has reached its limits,” said Jensen Huang. “We need new approaches to computing that can continuously scale, so that we can do more computing sustainably while continuing to drive down the cost of computing.”

He affirmed that the era of AI accelerators and related infrastructure has begun, saying that accelerated computing will bring about dramatic speed improvements compared to general-purpose computing across all industries.

Jensen Huang said, “In the future, data centers will be considered AI factories,” adding, “Their goal is to generate revenue, and in this case, to generate intelligence.”rdquo; said.

Jensen Huang introduced the NVIDIA Blackwell platform, which will provide a major upgrade to the world's AI infrastructure by implementing real-time generative AI on large-scale language models (LLMs) consisting of trillions of parameters.

Jensen Huang's one remark, which simplifies the previous explanations, is that "a larger GPU is needed." The Blackwell platform was created to address this challenge. Jensen Huang pulled a Blackwell chip out of his pocket and held it up next to the Hopper chip, dwarfing the latter.
The Blackwell platform enables enterprises to build and run real-time generative AI on large-scale language models (LLMs) at up to 25x less cost and energy than previous generations.

The Blackwell architecture GPU features 208 billion transistors. It is manufactured using a custom, two-reticle, 4NP TSMC process, forming a single, integrated GPU with a 10-terabyte-per-second (TB-per-second) chip-to-chip link.

Blackwell delivers 2.5x the performance per chip in FP8 for training and 5x the performance in FP4 for inference compared to the previous generation. It features fifth-generation NVLink interconnect that is twice as fast as Hopper and can scale up to 576 GPUs.

In addition, the innovative technologies of Blackwell architecture were mentioned as follows: △2nd generation transformer engine △5th generation NVLink △RAS engine △security AI △decompression engine.

"The amount of energy we save, the amount of networking bandwidth we save, the amount of time we save is enormous," Huang said. "This is a completely new industry because generative computing is the future. Our computing approach is fundamentally different. We've built a processor for the era of generative AI."

Nvidia boasted that the industry has already embraced Blackwell. According to the Blackwell announcement press release, Alphabet and Google CEO Sundar Pichai, Amazon CEO Andy Jassy, Dell Technologies CEO Michael Dell, Google DeepMind CEO Demis Hassabis, Meta CEO Mark Zuckerberg, Microsoft CEO Satya Nadella, OpenAI CEO Sam Altman, Oracle Chairman Larry Ellison, and Tesla and xAI CEO Elon Musk have expressed their support for Blackwell.

Blackwell will be deployed by all major global cloud service providers, leading AI companies, systems and server vendors, regional cloud service providers worldwide, and telecommunications companies.

"The entire industry is gearing up for Blackwell," said Jensen Huang. "Blackwell will be the most successful launch in NVIDIA's history."