Microsoft is working on its own artificial intelligence chips to train large language models and reduce reliance on Nvidia. The Information reports.
According to the publication, development of the processor began in 2019. Some Microsoft and OpenAI employees already have access to them to test their work with language models such as GPT-4.
Currently, Nvidia is the key supplier of AI server chips. According to оценкам from OpenAI, commercialising the ChatGPT chatbot requires more than 30,000 Nvidia A100 GPUs.
High demand for accelerators is driving shortages and high prices. The latest Nvidia H100 graphics processors are selling on eBay for more than $40 000.
To meet internal demand, Microsoft has stepped up work on the Athena project. Sources say they do not know whether the tech giant will use its own chips in the Azure cloud infrastructure. However, the company plans to make them available within Microsoft and OpenAI as early as next year.
There is also said to be a roadmap for chips that includes several future generations.
Sources emphasised that Athena is not a direct replacement for Nvidia. But the internal efforts could substantially reduce costs as Microsoft continues to embed AI-powered features into its products.
Earlier reports said that developers of AI systems were facing a shortage of servers.
In April, Google unveiled its AI supercomputer TPU v4. The company said it is 1.7 times faster and 1.9 times more energy-efficient than rivals built on Nvidia A100 chips.
In November 2022 Intel introduced server processors and accelerators for high-performance computing and AI workloads.
In January Meta announced plans to build the world’s largest AI supercomputer.