AMD executives say the company is taking an open-source approach as it takes on Nvidia in the chip wars.
Ramin Lone, AMD's corporate vice president of AI product management, spoke about the GPU shortage and the company's efforts onstage at Reuters' Momentum AI conference in San Jose, California on Tuesday.
Both AMD and Nvidia make GPUs (graphics processing units), which are used in video games and AI development because of their ability to handle large amounts of calculations. Nvidia controls more than 70% of the AI chip market, according to estimates by Mizuho Securities, and has major customers including Meta, Google, Amazon and OpenAI.
Demand for powerful chips to grow generative AI and train models is outstripping supply. Loan said GPU availability is a big issue in the chip industry, and AMD is “putting all of its GPUs in the hands of [it] You can make it right now.”
“In terms of vendor lock-in, we know what's going on right now: You have one very powerful vendor with completely closed, locked-in software,” Roane said, referring to Nvidia's CUDA computing platform, which helps developers build applications using Nvidia's GPUs.
Released in 2006, CUDA is only compatible with Nvidia's GPUs. The company's initial development of CUDA established a de facto industry standard, giving it a competitive advantage, but alternative programming models have struggled to catch on.
AMD has taken a different approach with its open-source ROCm software for GPU programming, Roane said. Because ROCm is available as open source, anyone can use, download or modify it for free.
“We're taking the opposite direction and our software, called ROCm, is completely open,” Roan said.
The tides are turning. Meta and Microsoft announced plans last year to buy AMD chips and move away from reliance on Nvidia's graphics processors. The AMD vice president added that customers are converting their CUDA programs to HIP, ROCm's programming language that's compatible with other GPUs. Microsoft is now introducing OpenAI's GPT-4 AI model into its AMD supplies, Roane said.
Alvin Nguyen, a senior analyst at Forrester, said AMD's open-source approach could help it gain market support.
“Their open source approach makes sense because creating a community with minimal barriers to adoption is their biggest opportunity to grow market share,” Nguyen told Business Insider. “It requires less support on their end compared to offering an in-house developed solution, and it helps keep costs down.”
This approach can keep costs down, but offering CUDA compatibility so customers can switch to other GPU suppliers still requires ongoing support from AMD to make it worthwhile, Nguyen said.
“The problem is that changes to CUDA could cause compatibility issues that will take time to resolve and leave users of non-NVIDIA products frustrated,” Nguyen said.
