An AI software business reported Friday that Advanced Micro Devices (AMD.O) chips are 80% as fast as Nvidia Corp (NVDA.O) chips, with a roadmap to matching their performance.
Nvidia controls the market for powerful chips required to produce ChatGPT and other AI applications that have swept the technology industry in recent months. The popularity of those services has lifted Nvidia’s worth beyond $1 trillion and caused a chip shortage that Nvidia is seeking to fix.
Tech businesses are seeking alternatives, hoping AMD will be a formidable competitor. MosaicML, an AI startup acquired for $1.3 billion this week, tested AMD and Nvidia AI CPUs.
MosaicML tested the AMD MI250 and Nvidia A100, one generation behind their flagship CPUs, but they are still popular.
Thanks to a new version of AMD software released late last year and a new version of Meta Platforms (META.O)-backed open-source software called PyTorch released in March, MosaicML showed AMD’s chip could reach 80% of Nvidia’s performance.
MosaicML’s CTO, Hanlin Tang, indicated that AMD’s upcoming software improvements should enable its MI250 chip rival Nvidia’s A100.
“For most (machine learning) chip companies out there, the software is the Achilles heel of it,” Tang said, adding that AMD has not paid MosaicML to study. AMD excels in software.
Tang claimed MosaicML uses PyTorch and AMD technologies to train a big language model without changing its code. “You can already switch to these today they’re essentially interchangeable” with Nvidia processors, Tang added.
MosaicML delivers software that helps organizations build AI systems in their data centers instead of paying OpenAI or ChatGPT for access. The company researched to show its clients had chip options outside Nvidia.
“Mosaic’s results reinforce our strategy of supporting an open and easy to implement software ecosystem for AI training and inference on AMD hardware,” AMD said, adding that it would continue collaborating with the startup to tweak its software.
