1 min read

NVIDIA’s CEO Delivers First DGX H200 to OpenAI

NVIDIA CEO Jensen Huang personally handed over the first NVIDIA DGX H200 to OpenAI, marking a significant milestone for both companies. In a shared post by Greg Brockman, president of OpenAI, Huang is seen alongside Brockman and chief Sam Altman, with the DGX H200 showcased prominently.

Brockman’s post on X celebrated the occasion, highlighting the dedication of the DGX H200 “to advance AI, computing, and humanity.” This gesture from Huang reflects a tradition of sorts, as he had previously donated the first DGX-1 AI supercomputer to OpenAI back in 2016, emphasizing his commitment to democratizing AI technology. Notably, Elon Musk was the recipient of the first DGX-1 at that time.

The introduction of the DGX H200 last year by NVIDIA represents a significant upgrade from its predecessor, the highly acclaimed H100. The H200 boasts 1.4 times more memory bandwidth and 1.8 times more memory capacity, enhancing its ability to handle demanding generative AI tasks effectively.

One of the key enhancements in the H200 is its faster memory specification, known as HBM3e, which elevates its memory bandwidth to 4.8 terabytes per second compared to the H100’s 3.35 terabytes per second. Additionally, the total memory capacity has been increased to 141GB, a significant jump from the 80GB offered by its predecessor.

Ian Buck, vice president of hyperscale and HPC at NVIDIA, emphasized the importance of efficient data processing at high speeds for creating intelligence with generative AI and HPC applications. He stated that with the NVIDIA H200, the industry’s leading AI supercomputing platform has become even faster, enabling solutions for some of the world’s most critical challenges.

In addition to delivering the DGX H200, NVIDIA also unveiled a new AI supercomputer featuring this advanced GPU. The NVIDIA DGX H200 leverages NVLink interconnect technology and the NVLink Switch System, combining 256 H200 superchips into a single GPU unit. This configuration achieves an impressive 1 exaflop of performance and offers 144 terabytes of shared memory, representing a substantial leap from the previous generation NVIDIA DGX A100 introduced in 2020.

The launch of the DGX H200 and the accompanying AI supercomputer underscores NVIDIA’s commitment to driving innovation in AI and HPC. By delivering cutting-edge technology to organizations like OpenAI, NVIDIA aims to empower them to tackle complex challenges and drive advancements in AI research and development.

Overall, the introduction of the NVIDIA DGX H200 marks a significant milestone in the field of AI computing, offering enhanced performance and capabilities to organizations at the forefront of AI innovation like OpenAI.

Leave a Reply