- An AMD executive said the firm embraces open source to tackle GPU shortages and challenge Nvidia.
- Nvidia controls over 70% of the AI chip market, leveraging its closed CUDA software.
- Meta and Microsoft are shifting to AMD chips.
An AMD executive said his company is taking an open-source approach as it challenges Nvidia in the chip wars.
Ramine Roane, corporate vice president of AI product management at AMD, addressed the GPU shortage and the company's approach onstage at Reuters' Momentum AI conference in San Jose, California, on Tuesday.
Both AMD and Nvidia produce GPUs, or graphics processing units, used in video gaming and AI development due to their ability to handle large amounts of computation. Nvidia controls more than 70% of the AI chips market, according to a Mizuho Securities estimate, and it has major customers such as Meta, Google, Amazon, and OpenAI.
The demand for generative AI and powerful chips to grow training models has outpaced supplies. Roane said that the availability of GPUs is a big problem within the chips industry and that AMD is "sending all the GPUs [it] can make right now."
"In terms of vendor lock-in, we know what's going on right now. There is one very strong vendor with the software that's completely closed and locked in," Roane said, referring to Nvidia's CUDA computing platform, which helps developers build applications using Nvidia's GPUs.
Launched in 2006, CUDA is only compatible with Nvidia's GPUs. The company's early development of CUDA effectively set an industry standard that gave it a competitive advantage, while alternative programming models have struggled to gain traction.
Roane says AMD took a different approach with its open-source ROCm software for GPU programming. Since ROCm is available as open source, it's free for anyone to use, download, or modify.
"We went the other way, and our software called ROCm is completely open," Roane said.
Tides are shifting. Meta and Microsoft announced plans last year to purchase AMD chips, moving away from a dependence on Nvidia's graphic processors. The AMD vice president added that customers are converting CUDA programs to HIP, ROCm's programming language, which can be compatible with other GPUs. Microsoft is now deploying OpenAI's GPT-4 AI model on AMD's supplies, Roane said.
Alvin Nguyen, senior analyst at Forrester, noted that AMD's open-source approach could help it gain market favor.
"Their open-source approach makes sense since creating a community with minimal barriers to adoption is their best chance at getting more market share," Nguyen told Business Insider. "It requires less support on their end, compared to providing an internally developed solution and will help keep costs low."
While this approach could keep costs down, Nguyen said that providing compatibility with CUDA to allow customers to convert to other GPU suppliers will still require continual support from AMD to make it worthwhile.
"The problem is that changes to CUDA can create compatibility issues that will take time to resolve but will leave users of non-NVIDIA products frustrated," Nguyen said.