OpenAI receives the world’s most powerful AI GPU from Nvidia CEO

The DGX H 200 is a successor to the firm’s H100, an AI supercomputer optimized for large generative AI and other transformer-based workloads.

The development also signifies a collaboration between two major forces in the AI industry: Nvidia concentrates on hardware advancements, while OpenAI specializes in software development.

First @NVIDIA DGX H200 in the world, hand-delivered to OpenAI and dedicated by Jensen "to advance AI, computing, and humanity": pic.twitter.com/rEJu7OTNGT

— Greg Brockman (@gdb) April 24, 2024

The next-gen AI processor

Nvidia’s unveiling of the DGX H200 marks a significant leap forward in high-performance computing. It boasts substantial improvements over its predecessor, the H100.

The platform, built on the Nvidia HopperTM architecture, has an Nvidia H200 Tensor Core GPU with enhanced memory to manage large volumes of data for high-performance computing and generative AI tasks.

The Nvidia H200 is the first GPU to support HBM3e, which is faster and more memory-efficient. This enables the advancement of scientific computing for HPC workloads, generative AI, and huge language models.

The standout upgrades over the H100 include a remarkable 1.4x surge in memory bandwidth and an impressive 1.8x increase in memory capacity. These enhancements translate to a staggering bandwidth of 4.8 terabytes per second and a capacious memory capacity of 141 GB.

According to Nvidia, such enhancements are pivotal for tackling the demands of training larger and more intricate AI models, particularly those deployed in generative AI applications that generate diverse content types like text, images, and predictive analytics.

Furthering the development of GPT-5

An effective data center architecture is essential for training AI models with hundreds of billions of parameters. This includes maximizing throughput, reducing server downtime rates, and utilizing multi-GPU clusters for tasks requiring a lot of computation.

“To create intelligence with generative AI and HPC applications, vast amounts of data must be efficiently processed at high speed using large, fast GPU memory,” With Nvidia H200, the industry’s leading end-to-end AI supercomputing platform, just got faster to solve some of the world’s most important challenges,” said Ian Buck, vice president of hyperscale and HPC at Nvidia.

These capabilities will advance OpenAI’s efforts to train GPT-5, which is touted to feature an evolved version of AI known as AGI. For reference, GPT-4 was trained on around 25,000 Nvidia A100 GPUs for about 100 days.

It is anticipated that GPT-5 will be a multimodal model with textual and visual inputs and outputs. AGI is anticipated to merge the functionalities of natural language processing, image recognition, data from sensors like cameras and microphones, pattern recognition, predictive analysis, abstraction, and logical reasoning.

The resultant AI system would be able to execute tasks akin to humans. Present-day AI tools like Siri, Alexa, Watson, and ChatGPT are classified as Artificial Narrow Intelligences (ANIs).

Although GPT-5’s release date is unknown, it’s safe to assume it’s in the works. (OpenAI worked on GPT-4 for at least two years before its formal release.) As with GPT-3.5, it’s also feasible that OpenAI will release an interim GPT-4.5 ahead of GPT-5.

OpenAI has not yet made an official announcement regarding the release schedule, but it is safe to say that the new Nvidia GPUs are set to advance the timelines.