Microsoft Secretly Developing AI Chips for Efficient Large Language Model Training
Reportedly, Microsoft has been working on its own AI chips since 2019, which can be utilized for training large language models, reducing the expenses of relying on Nvidia. According to The Information, these chips have been developed secretly, and some Microsoft and OpenAI staff members have been testing them for their efficacy in the latest large language models like GPT-4.
Microsoft has constructed a supercomputer to aid OpenAI, an AI research startup, in training large sets of models. To support ChatGPT and Bing AI chatbot, the supercomputer relies on thousands of NVIDIA A100 graphics chips. In 2019, Microsoft invested $1 billion in OpenAI to build a state-of-the-art supercomputer on a massive scale.
At present, Nvidia dominates the market as the main supplier of AI server chips, with many companies competing to purchase these chips. To commercialize ChatGPT, OpenAI is estimated to require over 30,000 of Nvidia’s A100 GPUs. The high demand for such high-end chips that can facilitate AI software deployment is evident from the fact that Nvidia’s latest H100 GPUs are selling for over $40,000 on eBay.
Nvidia is striving to construct as many chips as possible to cater to the demand, whereas Microsoft is apparently focusing on in-house development to save costs on its AI endeavors. The company has reportedly expedited its work on codename Athena, a venture aimed at developing its own AI chips. While it remains uncertain whether Microsoft will make these chips accessible to its Azure cloud clients, the tech giant is allegedly contemplating making its AI chips more widely available within Microsoft and OpenAI as early as next year. Furthermore, Microsoft has purportedly drafted a plan for the chips that encompasses several future generations.
1 Comment
Pingback: Embracing New Productivity Tools: The End of Cortana Support in Windows