jensen fan wants Nvidia’s co-founder and CEO said today at Computex in Taipei that it will bring generative AI to all its data centers. In his first public speech in almost four years, Huang made a number of announcements, including the release date of the chip, the DGX GH200 supercomputer, and partnerships with major companies, he said. Here’s all the news from the two-hour keynote.
1. Nvidia’s GForce RTX 4080 Ti GPUs for gamers are now in full production, working with partners in Taiwan to produce them in ‘volume’.
2. Huang announced Nvidia Avatar Cloud Engine (ACE) for games, a customizable AI model foundry service with pre-trained models for game developers. You will give your NPCs even more personality through AI-powered verbal interactions.
3. The Nvidia Cuda computing model currently serves 4 million developers and over 3,000 applications. Cuda had his 40 million downloads last year alone, including 25 million.
4. Mass production of the GPU server HGX H100 has started and is manufactured by “companies all over Taiwan,” Huang said. He added that this is the world’s first computer with a transformer engine.
5. Huang referred to Nvidia’s $6.9 billion acquisition of supercomputer chip maker Mellanox in 2019 as “one of the greatest strategic decisions” it has ever made.
6. Production of the next generation Hopper GPUs will begin in August 2024, exactly two years after the start of production of the first generation.
7. Nvidia’s GH200 Grace Hopper is now in full production. This superchip powers 4 PetaFIOPS TE, 72 Arm CPUs connected by chip-to-chip links, 96GB HBM3, and 576 GPU memory. “This is a computer, not a chip,” Huang said of the world’s first accelerated computing processor, which also has a huge memory. Designed for highly resilient data center applications.
8. If Grace Hopper’s memory isn’t enough, Nvidia has a solution: the DGX GH200. This is created by first connecting 8 Grace Hoppers to 3 NVLINK switches and then connecting the pods at 900GB. And finally, 32 are combined in another switch layer for a total of 256 Grace Hopper chips. The resulting ExaFLOPS Transformer Engine has 144 TB of GPU memory and acts as a giant GPU. Huang said Grace Hopper is so fast that it can run the 5G stack in software. Google Cloud, Meta, and Microsoft will be the first companies to have access to the DGX GH200 and conduct research into its capabilities.
9. Nvidia and SoftBank have formed a partnership to deploy Grace Hopper superchips in SoftBank’s new distributed data centers in Japan. You can now host generative AI and wireless applications on a multi-tenant common server platform, reducing costs and energy.
10. The SoftBank and Nvidia partnership is based on the Nvidia MGX reference architecture currently used in partnership with Taiwanese companies. It provides system manufacturers with a modular reference architecture to build over 100 server variations for AI, accelerated computing, and omniverse applications. Companies participating in the partnership include ASRock Rack, Asus, Gigabyte, Pegatron, QCT and Supermicro.
11. Huang unveiled the Spectrum-X high-speed networking platform for increasing the speed of Ethernet-based clouds. This includes a Spectrum 4 switch with 128 ports at 400GB/s and 51.2T/s. According to Huang, the switch is designed to enable a new breed of Ethernet and is designed end-to-end for adaptive routing, performance isolation and in-fabric computing. Also included is his Bluefield 3 Smart Nic that connects to a Spectrum 4 switch to perform congestion control.
12. WPP, the world’s largest advertising agency, has partnered with Nvidia to develop a content engine based on the Nvidia Omniverse. You will be able to create photo and video content for use in advertising.
13. Robot Platform Nvidia Isaac ARM is now available to anyone wanting to build a robot, full stack from chip to sensor. Huang said Isaac ARM is the first full reference stack for robotics, starting with a chip called Nova Orin.
Thanks in large part to its importance in AI computing, Nvidia’s stock price has skyrocketed over the past year to a current market valuation of around $960 billion, making it one of the world’s most valuable companies (Apple , Microsoft, Saudi Arabia only), Aramco, Alphabet and Amazon rank higher).
Chinese business in trouble
Chinese AI companies are no doubt keeping a close eye on the cutting-edge silicon that Nvidia puts out. On the other hand, they probably fear another round of US chip bans that could undermine progress in generative AI, which requires far more computing power and data than previous generations of AI. .
The US government last year restricted Nvidia from selling its A100 and H100 graphics processors to China. Both chips are used for training large language models such as OpenAI’s GPT-4. The H100, the latest generation chip based on his Nvidia Hopper GPU computing architecture with Transformer Engine inside, is expected to see particularly strong demand. Compared to A100, H100 can deliver 9x faster AI training and up to 30x faster AI inference on LLM.
China is clearly too big a market to miss. The chip export ban would cost NVIDIA an estimated $400 million in potential revenue in the third quarter of last year alone. So NVIDIA has resorted to selling low-speed chips to China that meet U.S. export control regulations. Longer term, however, China will likely seek stronger alternatives, and the ban will serve as a poignant reminder for China to achieve independence in key technology areas.
In a recent interview with the Financial Times, Huang said: [China] …We can’t buy it from the US, so we just make it ourselves. So the US should be careful. China is a very important market for the technology industry. ”
