ChatGPT maker OpenAI has received the fastest AI processor yet by Nvidia, just as it prepares for GPT-5 and Sora release.
OpenAI Receives NVIDIA DGX H200
On Wednesday, the president of Nvidia Jensen Huang personally hand-delivered the first “NVIDIA DGX H200” in the world to OpenAI. The AI company’s executives, Greg Brockman and Sam Altman were seen posing with the AI processor, together with Huang.
DGX H200 comes as the successor to Nvidia’s highly sought-after H100. Nvidia says the new processor, which will be made widely available later this year, is the “most efficient large memory supercomputer” yet.
The supercomputer is equipped with 32 NVIDIA Grace Hopper Superchips, interconnected with NVIDIA NVLink, according to reports. It also has a shared GPU memory space of 19.5TB and up to 900 gigabytes per second (GB/s) GPU-to-GPU bandwidth.
With GH200, developers can train huge AI models in weeks instead of months, Nvidia said. The proposed capabilities of the processor could switch up things for developers like OpenAI, which are working on larger models that consequently demand massive computing power.
OpenAI is Training Heavy Models
During Sam Altman’s discussion with Bill Gates in January 2024, he confirmed that GPT-5 is already under training, although the release date remains unknown.
GPT-5 is likely to support “multimodality,” which refers to models that can receive and produce not just texts but also images, speech, and video. Sam also said the ability to reason is another key focus for the company.
“Maybe the most important areas of progress will be around reasoning ability,” Sam said on Unconfuse Me. “Right now, GPT-4 can reason in only extremely limited ways.”
Another big launch in the pipeline from OpenAI is Sora, which was unveiled in February. Sora is an AI application that turns texts into videos with results so stunning that it had many predicting it could disrupt the content creation space.
Sora will be made available to the public later this year, according to reports.
All the models require a lot of computing power to pull off. Therefore, the GH200 comes as a big addition to OpenAI’s arsenal. Sam has also previously revealed plans to raise up to $7 trillion to enable OpenAI to start making its AI chips.