Software Company Predicts AMD AI Chips to be Equivalent to Nvidia’s
According to a report by an AI software firm on Friday, Advanced Micro Devices’ artificial intelligence chips are approximately 80% as fast as Nvidia Corp’s chips, and they have the potential to reach a similar level of performance in the future.
Nvidia dominates the market for powerful chips used to power ChatGPT and other AI services that have swept the tech industry in recent months. The popularity of these services has boosted Nvidia’s value to more than $1 trillion and led to a shortage of its chips, which Nvidia says it is working to address.
But in the meantime, tech companies are looking for alternatives, hoping that AMD will be a strong contender. That prompted MosaicML, an AI startup acquired for $1.3 billion earlier this week, to run a test comparing AI chips from AMD and Nvidia.
MosaicML evaluates the AMD MI250 and Nvidia A100, both of which are one generation behind each company’s flagship chips, but still in high demand.
MosaicML found that AMD’s chip could get 80% of the performance of Nvidia’s chip, thanks in large part to a new AMD software version released late last year and a new open-source software version supported by Meta Platforms called PyTorch, which was released in March.
Hanlin Tang, CTO of MosaicML, said the company believes future software updates from AMD should help its MI250 chip match the performance of Nvidia’s A100.
“For most (machine learning) chip companies, software is their Achilles’ heel,” Tang said, adding that AMD had not paid MosaicML to conduct its research. “Where AMD has been really successful is on the software side.”
Tang said MosaicML used its tools, PyTorch and AMD software, to train a large language model without making any changes to its code base. If developers can find AMD chips at the right price, “you can switch to these right now, they’re essentially interchangeable” with Nvidia chips, Tang said.
MosaicML sells software that makes it easier for companies to build AI systems in their own data centers, rather than paying for access to those systems from providers such as OpenAI, creator of ChatGPT. The company said it conducted the research to show its customers have chip options beyond Nvidia.
“Mosaic’s results reinforce our strategy to support an open and easy-to-implement software ecosystem for AI training and inference on AMD hardware,” AMD said in a statement, adding that it will continue to work with the company to tune its software.
