Nvidia CEO: Latest Generation Chips Fully Produced, AI Performance Increased Fivefold

JAKARTA - Nvidia CEO Jensen Huang stated that Nvidia's latest generation chips have now entered full production. The new chip is claimed to be able to deliver artificial intelligence (AI) computing performance up to five times higher than the previous generation, especially for running chatbots and various AI applications.

The statement was made by Huang in his speech at the Consumer Electronics Show (CES) in Las Vegas. He revealed that Nvidia's latest chips are scheduled to launch later this year and are currently in Nvidia's labs to be tested by a number of AI companies. This move comes amid Nvidia's increasing competition with traditional rivals and large customers who are beginning to develop their own AI chips.

Nvidia's latest chip platform is called Vera Rubin. The platform consists of six separate chips, with a flagship server carrying 72 units of graphics processing units (GPUs) and 36 new-generation central processing units (CPUs). Huang explained that the system can be strung together in a single unit or "pod" containing more than 1,000 Rubin chips, which are capable of increasing the efficiency of AI "token" creation by up to 10 times.

To achieve this performance leap, Nvidia uses a proprietary type of data that is expected to be adopted more widely by the industry. Huang said this performance improvement is very significant even though the number of transistors only increased by about 1.6 times compared to the previous generation.

Although Nvidia still dominates the market for training AI models, competition is tightening at the stage of delivering AI services to hundreds of millions of users. Competitors come from companies such as AMD, as well as large customers such as Google which are also developing internal AI chips. In response to this challenge, Nvidia introduced a new technology called context memory storage, which is designed to help chatbots provide faster responses in long and complex conversations.

In addition to computing chips, Nvidia also showcased the latest generation of network switches with co-packaged optics technology. This technology is important for connecting thousands of machines in a single integrated system and will compete directly with solutions from Broadcom and Cisco.

Nvidia said CoreWeave will be one of the first companies to use the Vera Rubin system. In addition, Microsoft, Oracle, Amazon, and Alphabet are also expected to adopt the technology.

In another announcement, Huang highlighted new software for autonomous vehicles that helps the system make route decisions while providing a data trail for engineers. The software, called Alpamayo, will be released more broadly, complete with open-source training data so that automotive manufacturers can evaluate and trust the results.

Huang also responded to the acquisition of talent and chip technology from startup Groq last month. According to him, this move will not affect Nvidia's core business, but has the potential to give birth to new products in the future.

On the other hand, Nvidia wants to show that its latest chips are able to surpass the performance of old chips such as the H200, which are still in demand in the Chinese market. Huang admitted that demand for the H200 in China remained strong. Nvidia CFO Colette Kress added that the company had applied for permission to ship the chips and was still awaiting approval from the US government and other relevant countries.