Who is Jensn Huang?
Huang, considered by many to be a sort of new Steve Jobs, from the stage in Taipei said he believes that these innovations will not only facilitate new business models, but will also significantly improve the efficiency of existing models in a multitude of sectors. One of the highlights of the keynote was the launch of Grace Hopper, a platform that combines the power of the energy-efficient Nvidia Grace CPU with the high-performance Nvidia H100 Tensor Core GPU. This all-in-one module enables enterprises to achieve unprecedented AI performance. In addition, Huang introduced the DGX GH200, an AI supercomputer with large memory capacities that can integrate up to 256 Nvidia Grace Hopper Superchips into a single datacenter-sized GPU. In fact, it’s a reference model that high-ranking Nvidia customers like Meta, Microsoft, and Google will be able to use for their AI projects.
How the Dgx Gh200 works
With one exaflop of performance and 144 terabytes of shared memory, the DGX GH200 outperforms its predecessors nearly 500 times, enabling developers to build advanced chatbot language models, algorithms, and sophisticated neural networks for tasks such as fraud detection and data analysis. The CEO emphasized that “DGX GH200 AI supercomputers integrate Nvidia’s most advanced accelerated computing and networking technologies, pushing the boundaries of AI further”. Huang also unveiled the Nvidia Avatar Cloud Engine (ACE) for games, a service that allows developers to build and deploy custom AI models for speech, conversation and animation. ACE equips non-playable characters with conversational skills, allowing them to answer questions with realistic evolving personalities. The toolkit includes essential AI base models, such as Nvidia Riva for speech detection and transcription, Nvidia NeMo for generating custom responses, and Nvidia Omniverse Audio2Face for animating those responses. Additionally, Nvidia announced its partnership with Microsoft to lead innovation in the era of generative AI for Windows PCs. The partnership includes the development of advanced tools, frameworks and drivers that streamline the process of developing and deploying AI on PCs. The collaboration aims to improve and expand the installed base of over 100 million PCs with RTX GPUs equipped with Tensor Cores, thereby increasing the performance of over 400 AI-accelerated Windows applications and games.
Leveraging Generative AI for Digital Advertising
In Taipei, there was also a way to understand that the potential of generative AI can expand to the digital advertising sector, where Nvidia is already present with the company WPP. Together, the companies have developed an innovative content engine on the Omniverse Cloud platform. This allows creative teams to connect their 3D design tools, such as Adobe Substance 3D, to create product digital twins within Nvidia Omniverse. Using AI tools trained on data from responsible sources and powered by Nvidia Picasso, workers will be able to rapidly produce virtual sets, to generate a multitude of ads, videos and 3D experiences customized for global markets, accessible on any web device.Going Further, Huang pointed out that by leveraging Nvidia technologies, electronics manufacturers such as Foxconn Industrial Internet, Innodisk, Pegatron, Quanta and Wistron are transitioning to digital workflows, bringing the vision of all-digital smart factories closer to reality. According to Huang: “The world‘s largest industries create physical products. But by trying them digitally first, they could save billions of dollars.” The integration of Omniverse and generative APIs has enabled these companies to make connections between design and manufacturing tools, thereby building digital replicas of their factories. The latest addition, Nvidia Metropolis for Factories, enables the creation of custom quality control systems, giving manufacturers a competitive edge and enabling them to develop cutting-edge AI applications.
A new range of AI supercomputers
Nvidia Helios is instead the AI supercomputer that should be operational by the end of the year. It will use four DGX GH200 systems interconnected with Nvidia’s Quantum-2 InfiniBand network, delivering bandwidth up to 400 Gb/s. As a result, data throughput for training large-scale AI models will be greatly improved. In addition to this groundbreaking development, Nvidia has introduced Nvidia MGX, a modular reference architecture that allows developers to efficiently and cost-effectively build different server configurations tailored to AI, HPC and Nvidia Omniverse applications. With the MGX architecture, manufacturers can develop standardized CPUs and accelerated servers, using modular components. These configurations support a range of GPUs, CPUs, data processing units (DPUs), and network adapters, including x86 and Arm processors. Additionally, MGX configurations can be housed in both air- and liquid-cooled chassis.