Ian Buck, VP of NVIDIA, introduces the importance of computing and NVIDIA's role in accelerating it to transform various scientific and industrial domains.

• Ian Buck, VP of HPC and Hyperscale Datacenter at NVIDIA, opens Supercomputing 2023.
• Computing is fundamental to modern civilization, driving innovation in scientific discovery, industrial automation, and AI.
• The stagnation of CPU performance growth has led to the search for new computing pathways.
• NVIDIA's accelerated computing serves as a multi-domain acceleration platform with architectural compatibility.

Accelerated computing significantly reduces system costs and energy usage, transforming industries such as automotive with NVIDIA's GPUs.

• Accelerated computing reduces system cost and energy use, as evidenced by Siemens and Mercedes' collaboration on the electric EQE vehicle.
• The use of NVIDIA Hopper H100 GPUs allowed for a 3x cost reduction and 4x energy consumption reduction, facilitating faster design iterations.

Accelerated computing is a full-stack problem requiring innovation in processors, libraries, and data center design, with AI transforming high-performance computing.

• Full-stack optimization involves world-class processors, domain-specific libraries, and close work with application developers.
• The data center is now the new unit of compute, with NVIDIA's software and hardware solutions scaling applications across data centers.
• AI as software that writes software is transforming high-performance computing, providing further acceleration in scientific domains.

NVIDIA's cuQuantum and CUDA Quantum are powering quantum computing research, with partnerships and global deployments advancing the field.

• Supercomputers are central to quantum computing research and are used to simulate future quantum processors and improve today's quantum computers.
• NVIDIA's platform enables researchers to control and correct quantum errors for optimal performance and integrate quantum processors into future supercomputers.
• Partnerships with quantum computing providers and the development of CUDA Quantum facilitate the era of quantum-accelerated supercomputing.

Generative AI is revolutionizing scientific research, enabling breakthroughs in chemistry, biology, and the study of genomic DNA sequences.

• Generative AI impacts various scientific domains, from climate and disease research to astronomy.
• Examples include the University of California Berkeley's chemistry assistant for MOFs and Northwestern University's DNABERT for genomic DNA sequences.
• Generative AI helps in identifying COVID-19 variants and has predictive power for experiments, supported by NVIDIA's array of generative AI technologies.

NVIDIA's range of solutions like Modulus, NeMo, and BioNeMo empower researchers with generative AI technologies for computational biology and drug discovery.

• NVIDIA Modulus merges physics-based causality with data for real-time predictions in engineering.
• NeMo is a multi-modal framework that facilitates large language model training and BioNeMo accelerates computational biology and drug discovery.
• NVIDIA's Science and Engineering Teaching Kit, developed with Brown University, includes the Modulus Framework for educational purposes.

AI Supercomputing requires world-class AI factories, with NVIDIA's Eos AI Supercomputer setting performance records in large language model training.

• AI accelerator breakthroughs necessitate AI factories for building, deploying, and maintaining scientific foundational AI models at scale.
• MLPerf benchmarks provide unbiased evaluations of training and inference performance, with NVIDIA's Eos system delivering record-setting results.
• Partnership with Microsoft Azure built an AI supercomputer in the cloud, setting a new scale record for large language model training.

NVIDIA's memory standards advancements with GPUs like Volta, Ampere, and Hopper lead to the H200 GPU with HBM3e memory, boosting performance for AI and HPC.

• NVIDIA has pioneered the integration of GPUs with the latest HBM2, 2E, and HBM3 memory technologies, enhancing computational power.
• The H200 GPU, featuring HBM3e memory, offers substantial increases in memory performance and capacity, accelerating AI and HPC applications.
• The HGX H200's compatibility with HGX H100 systems allows for easy system upgrades without the need for redesign.

The Grace Hopper Superchip bridges the gap for legacy CPU applications, offering energy efficiency and a coherent architecture for diverse scientific applications.

• The Grace Hopper Superchip features NVLink interconnect for coherent access to high-speed memory, enhancing energy efficiency for HPC.
• The GH200 Superchip is tailored for generative AI and HPC applications like AI chatbots and scientific simulations, offering transformative performance.
• The GH200 platform's architecture facilitates low-latency coupling of quantum computers, anticipating the era of quantum-accelerated supercomputing.

AI supercomputers powered by Grace Hopper will be brought online by various partners and organizations, contributing to the global AI ecosystem.

• Grace Hopper AI supercomputers, powered by the GH200 superchip, will be available through partners such as Dell, Lenovo, and cloud service providers.
• Early access systems have been purchased by major enterprises, including NASA and Total Energies, with NVIDIA LaunchPad providing early access to hardware and software.
• The new class of AI supercomputers, including the Alps, Venado, Vista, and Jupiter systems, will deliver 200 exaflops of AI performance for scientific advancements.