AI has become a ubiquitous theme in the tech industry and NVIDIA stands out as a key player, often referred to as the kingmaker of AI. As such, one can expect GTC 2024 to showcase groundbreaking advancements and applications in AI across various domains.
Our analysts are covering the event live as well as virtually. Here’s what Brady Wang, Associate Director at Counterpoint Research, has to say from the San Jose McEnery Convention Center:
NVIDIA is the hottest company in the world right now. The company’s focus on multimodal AI training in data centers, particularly emphasizing the energy-efficient GB200 systems and DGX SuperPODs with enhanced inference capability, will contribute significantly to its growing revenues in the Data Center sector. This transition is evident as companies move from training Large Language Models (LLMs) to multimodal AI models. Furthermore, NVIDIA is highlighting the integration of AI into the physical world, with industries like robotics, automotive, and extended reality (XR) incorporating AI into the 3D realm, unlocking new possibilities and use cases.
Decoding GTC 2024 – Analyst Takeaways on Key Announcements
NVIDIA Blackwell: World’s most powerful chip launched
NVIDIA introduced Blackwell, a powerful chip with 208 billion transistors, designed for training massive 10 trillion parameter AI models. It uses a custom 4NP TSMC process and connects GPU dies with a 10 TB/second chip-to-chip link. They also launched GB200 Grace Blackwell Superchip, linking B200 Tensor Core GPUs with the Grace CPU. Counterpoint Research predicts NVIDIA’s Data Center revenue to reach over $72 billion in 2024, a 134% increase from last year, thanks to Blackwell’s adoption by cloud providers and countries for their AI supercomputers.
NVIDIA NIM: Foundational software for AI deployments through API services
NVIDIA is driving AI adoption forward with the introduction of NVIDIA NIM, an API-based platform designed for AI inferencing. This is in addition to the NVIDIA Enterprise platform which already provides access to a suite of services for AI training and inferencing. This platform facilitates quicker deployment times and offers plug-and-play options, aiming to simplify the intricacies involved in developing software for AI inferencing. It provides access to ready-made platforms through APIs for inferencing AI models for real-time insights. NVIDIA has indicated a price of $1/per GPU/per hour for access to NVIDIA NIM.
Denmark next country for AI sovereignty – Supercomputer with 191 DGX systems
Denmark’s collaboration with Novo Nordisk and NVIDIA to establish an AI Innovation Center underscores the global shift towards domestic AI data centers, emphasizing data privacy and insights protection. This strategic initiative, encompassing healthcare, drug discovery, quantum, and social science, reflects a broader trend among countries. Denmark’s commitment, including the deployment of over 191 DGX systems, showcases a concerted effort towards advanced AI capabilities in key sectors.
NVIDIA enters the XR market
NVIDIA has introduced Ominverse Cloud APIs, enabling Apple Vision Pro users to personalize and share high-fidelity OpenUSD files on mobile devices seamlessly, leveraging cloud-compensated computing for optimal content and performance. This integration of NVIDIA Omniverse and Apple Vision Pro holds promise for a revolutionary mixed-reality (MR) experience, opening doors to diverse industrial, productivity and entertainment applications.
Partnership with BYD
The partnership is an extension of the existing partnership with BYD on the NVIDIA DRIVE platform which further enhances, optimizes and digitalizes the complete product lifecycle from vehicle concept to retail through the integration of AI and Metaverse technology.
Computational lithography accelerated by integrating cuLitho in production processes
The integration of NVIDIA’s cuLitho in semiconductor manufacturing will speed up photomask production that takes weeks to days and help reduce power needs and potential environmental harm, further enabling higher chip density and yield.
As semiconductor scaling becomes more complex, a greater number of foundries, EDA vendors and lithography equipment makers will be integrating the NVIDIA cuLitho in their process flow as the amount of data processing requirements in innovative mask techniques, photoresist modeling and high-NA EUV technology for new technology nodes is expected to increase significantly.
Physical manifestation of AI – Project GR00T
Project GR00T is a major step forward in humanoid robot development, showing NVIDIA’s commitment to pushing AI boundaries. Jetson Thor, a new computing system for humanoid robots, is a key part of this effort. With a powerful GPU based on NVIDIA’s latest architecture and a transformative engine delivering 800 teraflops of AI performance, Jetson Thor enables humanoid robots to run advanced AI models like GR00T, marking a significant milestone in humanoid robotics.
