The new network card is similar to Nvidia’s ConnectX-7 card.
Microsoft is reportedly developing a new network card that could enhance the performance of its Maia AI server chip and potentially reduce its reliance on chip designer Nvidia. The project is being led by Pradeep Sindhu, co-founder of Juniper Networks, who joined Microsoft following the acquisition of his server chip startup, Fungible, last year.
The new network card is said to be similar to Nvidia’s ConnectX-7 card, which is sold alongside Nvidia’s GPUs. If successful, the development of this equipment could take over a year and may significantly speed up the time required for OpenAI to train its models on Microsoft servers, making the process more cost-effective.
Microsoft has made major investments in OpenAI, the creator of ChatGPT, and has integrated its technology into various products, positioning itself as a key player in the artificial intelligence software market. In November, Microsoft introduced Maia, a chip designed to run large language models and support AI computing.
OpenAI CEO, Sam Altman feels that current chip producing capacity is not enough to withstand the ongoing rapid AI development. He is in talks with middle east investors and big fabs to foray into semiconductor manufacturing.
The Maia 100 is the inaugural model in the Maia series of accelerators designed specifically for artificial intelligence applications. With 105 billion transistors, the Maia 100 is one of the largest chips on 5-nanometer process technology.