Source: Securities Times Author: Zhou Chunmei

In response, some netizens ridiculed that it was actually “directly from Boss.” However, the reporter discovered that this is not the first time that Hwang In-hoon has personally delivered the latest supercomputer to OpenAI. Shortly after the establishment of OpenAI in 2016, Hwang In-hoon delivered the world's first DGX-1 to the OpenAI office, but the main recipient at the time was Musk, who had not left OpenAI.

Ultra high performance AI computer, equipped with H200 chip

In November of last year, Nvidia officially unveiled the H200 at the 2023 Global Supercomputing Conference (SC23). This is Nvidia's new image processing chip (GPU) in the field of artificial intelligence, and is positioned as another landmark product of Nvidia in the field of AI computing.

Compared to the previous main H100, this chip is also based on the Hopper architecture. The main upgrades include 141GB of HBM3e video memory. The video memory bandwidth was increased from 3.35Tb/s of the H100 to 4.8Tb/s, and the performance was increased by about 60% to 90%. Since Meta had not released Llama3 when the H200 was released, the H200 deduced on the Llama 2 large model with 70 billion parameters. The results showed that the inference speed was twice as fast as the H100, and the energy consumption in inference was directly reduced by half compared to the H100.

At the time, Nvidia said the H200 would be shipped in the second quarter of 2024, and Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle Cloud would be the first cloud service providers to deploy H200 instances. The reporter noticed that on March 28, foreign media reported that Nvidia's H200 had already begun to be supplied.

However, the stock price of Nvidia, which had been making great strides before, recently experienced a major correction. On April 19, local time, Nvidia's stock price plummeted by more than 10%, the biggest daily decline since March 2020. The market value evaporated by more than 200 billion US dollars (about 1448 billion yuan) in one day. However, the reason for this day's sharp decline was that Nvidia's server supplier Chaowei Computer was abnormal and failed to announce preliminary financial results in advance, causing investors to worry.

Ultramicrocomputer previously announced that it will release financial results for the third quarter of the 2024 fiscal year on April 30. Outsiders believe that this move may mean that its quarterly results are weaker than expected. In recent days, Nvidia's stock price has begun to fall again after two days of recovery. Some industry insiders analyzed that Nvidia's stock price will remain sideways to a negative decline until the ultra-micro computer announces its results.

Hwang In-hoon delivers supercomputers to OpenAI for the second time

In fact, this isn't the first time that Hwang In-hoon has delivered Nvidia's latest supercomputer to OpenAI. Shortly after OpenAI was founded in 2016, Hwang In-hoon delivered the world's first supercomputer DGX-1 equipped with 8 Nvidia P100 chips to OpenAI's office. However, in addition to Ultraman, the recipient at the time was Elon Musk, who had yet to leave OpenAI.

According to Hwang In-hoon's earlier recollection in an interview with the media, he delivered the world's first AI supercomputer to Musk. “It took us five years to build the DGX. It was initially built for our engineers, and I mentioned it at a conference, Musk saw it and said he wanted one, then he told me about OpenAI, so I delivered the world's first AI supercomputer to OpenAI.” Hwang In-hoon said.

For OpenAI, which was still in the early stages of entrepreneurship at the time, this supercomputer was undoubtedly a major “artifact”, greatly improving the speed of large-scale model development and training. On this DGX-1 supercomputer, Hwang In-hoon wrote, “For Elon and the OpenAI team,” while Musk, Ultraman, and other early employees also left their signatures below.

Musk was one of the original co-founders of OpenAI. He helped OpenAI recruit scientists and engineers, and provided an investment of around 50 million dollars. However, in 2018, Musk withdrew from the OpenAI board of directors on the grounds of a “conflict of interest.” On February 29 of this year, Musk also filed a lawsuit against OpenAI, the company's CEO Ultraman, and President Brockman in San Francisco, accusing OpenAI of violating its “original intention” and moving from an open source non-profit artificial intelligence company to a closed source, for-profit company, and demanding that OpenAI resume open source and pay compensation.

edit/lambor