Nvidia, open source suddenly!

On June 14 local time, Nvidia opened Nemotron-4 340B (340 billion parameters) series models.

According to Invista, developers can use this series of models to generate synthetic data, which can be used to train large language models (LLM) for business applications in healthcare, finance, manufacturing, retail and other industries. Nemotron-4 340B includes the basic model Base, instruction model Instrument and reward model Reward. Nvidia used 9 trillion tokens (text units) for training.

Nemotron-4 340B-Base performs common sense reasoning tasks, For example, in ARC-c, MMLU and BBH benchmark tests, it can compete with Llama-3 70B, Mixtral 8x22B and Qwen-2 72B models.

As of the closing of the US stock market on Friday, Nvidia, the first trading volume in the US stock market on Friday, closed 1.75% higher, hitting a new record high with a turnover of US $40.723 billion and a market value of US $3.244 trillion. Nvidia rose 9.1% this week, the eighth consecutive week of gains.

Huatai Securities issued a report stating that Nvidia split its shares in a "one for ten" way to increase liquidity, which took effect after the closing of last Friday (7th). The company's last announcement of "one split and four" was in 2021. From the announcement to implementation (from May 21 to July 19 of the same year), three months later (from July 19 to October 20 of the same year), and six months later (from July 19 of 2021 to January 20 of 2022), the stock price increased by 25%, 18%, and 29%, respectively.

As the leader of the AI industry, Nvidia's performance this year is impressive. Up to now, NVIDIA's share price has increased by more than 150% in total within the year, far exceeding the increase of the NASDAQ index over the same period.

The soaring share price of Nvidia was driven by product upgrading, outstanding performance and other favorable factors. Nvidia CEO Huang Renxun previously announced that the company plans to launch a high-performance version of Blackwell Ultra in 2025, and a new AI chip platform Rubin in 2026. Rubin's Ultra version will debut in 2027.

In addition, the first quarter financial report just released recently showed that NVIDIA's adjusted earnings per share reached 6.12 dollars, and its revenue reached 26 billion dollars, up 461% and 262% respectively compared with the same period last year. Among them, the revenue of data center business increased by 427% year on year, reaching US $22.6 billion, accounting for 86% of the company's total revenue.

Tesla CEO Musk said on the social platform on June 3 that his AI startup xAI would buy 300000 Nvidia B200 AI chips. "The next important step may be to have about 300000 B200 chips equipped with CX8 network next summer," Musk said.

According to the Daily Business News, for Nvidia, in the short term, technology companies need to constantly invest in computing power and resources, and chip demand continues to rise. MarketWatch has a view that NVIDIA GPU is essentially a new gold/oil in the field of technology, and it may become the world's first stock with a market value of $4 trillion.

However, in addition to the optimistic voice, some analysts said that no matter the closed source models such as ChatGPT, or the open source models such as Meta and Google, they have not yet explored a stable path to commercialization. From the perspective of price and market business model, as well as the fact that technology giants cluster to develop their own chips, NVIDIA's chip procurement in the future is also uncertain.

Daily economic news Comprehensive daily economic news, public news

Disclaimer: The content and data in this article are for reference only and do not constitute investment advice. Please verify before use. Operate accordingly at your own risk.