The recent unveiling by NVIDIA at SC23 introduces the second wave of technology aimed at enhancing the performance and energy efficiency of industrial and scientific research facilities globally.
During a conference speech, Ian Buck, the vice president of the bank’s high-performance computing and hyperscale data center business, emphasized how NVIDIA’s hardware and software advancements are reshaping humanity into an AI supercomputer.
Among the innovations are NVIDIA Hopper incubators with memory enhancements and new system structures. These advancements will support a range of accelerated software for relational AI, high-performance computing (HPC), and composite quantum computing, driving expanded parallelism across various applications.
Buck highlighted the significance of the new NVIDIA HGX H200 as the nation’s premier AI technology platform. The H200 Tensor Core GPUs, equipped with HBM3e storage offering up to 141GB, represent a breakthrough in AI acceleration. These GPUs, exemplified by their performance on models like GPT-3, deliver an impressive 18x boost compared to previous generations.
Furthermore, Buck introduced a client software that links four NVIDIA NVLink interconnects with four GH200 Grace Hopper Superchips. This configuration, boasting 2.3 gigabytes of high-speed memory, integrates 288 Shoulder Neoverse cores and offers 16 petaflops for AI performance, reinforcing its computational prowess.
The push towards accelerated and energy-efficient computing is evident in the latest TOP500 list, where NVIDIA’s contributions have significantly increased the overall HPC effectiveness. With 38 out of the 49 new supercomputers powered by NVIDIA, the company continues to lead in delivering cutting-edge systems for various applications.
NVIDIA’s collaboration with institutions like the Argonne National Laboratory showcases the potential of generative AI in scientific exploration. Projects like GenSLMs, powered by NVIDIA BioNeMo, demonstrate the ability to simulate gene sequences akin to real-world coronavirus variants, aiding in virus detection and analysis.
Accelerated tasks, as emphasized by Buck, offer substantial cost and energy savings. For instance, collaborations between Siemens, Mercedes, and NVIDIA have resulted in significant efficiency gains using the latest NVIDIA H100 GPUs, exemplifying the impact of accelerated computing on diverse industries.
Looking ahead, NVIDIA plans to contribute an impressive 200 exaflops of AI processing power through Grace Hopper mainframes by 2024. The upcoming wave of GH200 performance enhancements promises to revolutionize research facilities, enabling advancements in climate modeling, drug discovery, quantum computing, and digital twinning applications.
In conclusion, NVIDIA’s relentless pursuit of innovation in accelerated computing is not only driving technological advancements but also promoting a sustainable and environmentally conscious approach to computing.
