Nvidia said it plans to launch a new product designed to handle complex tasks such as video generation and software development, as its chips and systems are at the center of a boom in artificial intelligence computing. Nvidia said the product, called Rubin CPX, will be available by the end of 2026. It will take the form of a card that can be embedded into existing server computer designs, or used in standalone computers capable of running in parallel with other hardware in the data center.
The chipmaker said the design, an offshoot of its new Rubin product line launching next year, could make certain types of artificial intelligence work more efficient. This is yet another example of Nvidia maintaining an impressive pace in rolling out hardware and software that its competitors have yet to match, ensuring that the industry continues to rely on its products.
Nvidia said that some parts of the inference link in AI computing - the process of generating answers - are not as efficient as expected. The reason is that a single graphics processor is responsible for both understanding the input and generating and outputting responses. By separating the "understanding input" part from the "generating response" part and handing it over to new GPU chips, Nvidia says customers will get more efficient hardware.