NVIDIA's founder and CEO, Jensen Huang, announced a range of new systems, services, and software at the COMPUTEX conference in Taipei. These platforms are designed to help companies harness the potential of generative AI. As the most transformative technology of our time, AI is changing the way businesses operate across industries, from advertising and manufacturing to telecom. Huang's keynote marked his first in-person event since the of the pandemic.

During the event, Huang unveiled the DGX GH200, a large-memory AI supercomputer capable of combining up to 256 NVIDIA GH200 Grace Hopper Superchips. This enables developers to build complex algorithms through generative AI chatbots, recommender systems, and graph neural networks utilized for fraud detection and data analytics. The DGX GH200 packs an exaflop of performance and 144 terabytes of shared memory, making it an energy-efficient, high-performance solution.

The company is building its own AI supercomputer, NVIDIA Helios, set to be online this year. Helios will use DGX GH200 systems linked with NVIDIA Quantum-2 InfiniBand to deliver high-speed data throughput for training large AI models. Google Cloud, Meta, and Microsoft are among the first organizations expected to gain access to the system as a blueprint for future hyperscale generative AI infrastructure.

NVIDIA announced a modular reference architecture called NVIDIA MGX, aimed at building a broad range of AI, HPC, and NVIDIA Omniverse applications. More than 400 system configurations are coming to the market, powered by NVIDIA's latest Hopper, Grace, Ada Lovelace and BlueField architectures.

Moreover, system makers can use the MGX architecture to build over a hundred different server configurations, customizing them for specific needs quickly and cost-effectively. MGX helps manufacturers create CPU and servers that use modular components while supporting NVIDIA's full line of GPUs, CPUs, data processing units (DPUs), and network adapters.

The platforms will help SoftBank explore 5G applications in autonomous driving, AI factories, augmented and virtual reality, computer vision, and digital twins. The system will reduce costs while increasing spectral efficiency and reducing energy consumption.

In the computing world, networking is of paramount importance. NVIDIA Spectrum-X is a new networking platform announced during NVIDIA CEO Jensen Huang's keynote at the GPU Technology Conference (GTC) 2021. This platform promises to be an improvement on traditional Ethernet fabrics by delivering 1.7x gains in AI performance and power efficiency. The Spectrum-X platform combines Spectrum-4 Ethernet switches with BlueField-3 Data Processing Units (DPUs) and software to create a networking system that Huang says is built specifically for Ethernet-based AI clouds.

The technology is available now from system vendors like Lenovo, Supermicro and Dell Technologies. With a promise to improve efficiency and performance, the technology is expected to be seen as a critical innovation in the industry.

In addition, Jensen Huang also spoke of NVIDIA Avatar Cloud Engine (ACE) for Games, a foundry service that allows developers to build and deploy custom AI models for speech, conversation and animation. Huang said this would give non-playable characters new conversational skills with more advanced interactions allowing users to engage with them in a more intuitive way. The software uses NVIDIA Riva to detect and transcribe players' speech, before using NVIDIA NeMo to generate customized responses that are then animated with NVIDIA Omniverse Audio2Face. The result is NPCs with lifelike personalities and evolving traits, improving gamer engagement and interaction.

PC-developer tools, frameworks, and drivers have now made it easier to integrate AI. For example, the Microsoft Olive toolchain is used to optimize and deploy GPU-accelerated AI models, while new graphics drivers can improve DirectML performance on Windows PCs with NVIDIA GPUs. This collaboration is expected to improve the performance of more than 400 AI-accelerated Windows apps and games for an installed base of over 100 million PCs with Tensor Cores.

Generative AI has massive implications for the digital advertising industry, which is valued at an estimated $700 billion. WPP, the world’s largest marketing services organization, is working with NVIDIA to develop a generative AI-enabled content engine on Omniverse Cloud. With this engine, creative teams can produce virtual sets ready for advertising campaigns, videos and 3D experiences for global markets, using 3D design tools like Adobe Substance 3D and generative AI tools trained with NVIDIA Picasso.

Meanwhile, generative AI APIs with NVIDIA technologies are connecting design and manufacturing tools, facilitating the implementation of advanced AI techniques in more sectors. Pegatron and Foxconn Industrial Internet are using NVIDIA technology to simulate robots with Isaac Sim, Isaac AMR and Metropolis, resulting in improved production efficiencies, quality control, and throughput. These AI solutions make the manufacturing process more efficient. Watch full keynote here: https://youtu.be/i-wpzS9ZsCs