In a recent report by AI software firm MosaicML, Advanced Micro Devices’ (AMD) artificial intelligence chips were found to be approximately 80% as fast as Nvidia’s chips, with the potential to match their performance in the future. Nvidia currently dominates the market for powerful AI chips, which are essential for technologies like ChatGPT. However, the growing demand for AI services has caused a shortage of Nvidia chips, leading tech companies to seek alternatives.
To assess AMD’s potential as a strong competitor, MosaicML conducted a test comparing AMD’s MI250 chip with Nvidia’s A100. Although both chips are one generation behind the flagship models of their respective companies, they remain in high demand. MosaicML discovered that AMD’s chip could achieve 80% of Nvidia’s performance, primarily due to a new version of AMD software released in late 2022 and the availability of PyTorch, an open-source software supported by Meta Platforms, which was released in March 2023.
According to Hanlin Tang, MosaicML’s Chief Technology Officer, further software updates from AMD currently in development are expected to enable the MI250 chip to match the performance of Nvidia’s A100. Tang highlighted that software plays a critical role in the performance of machine learning chips, where AMD has excelled.
MosaicML utilized PyTorch and AMD software to train a large language model without needing any modifications to its code base. Tang suggested that developers who can find AMD’s chips at the right price can already consider them as interchangeable with Nvidia chips. He emphasized that AMD’s strength lies in its software capabilities.
MosaicML, recently acquired for $1.3 billion, offers software that simplifies the creation of AI systems within company data centers, enabling them to reduce reliance on providers such as OpenAI for access to AI capabilities like ChatGPT. The research conducted by MosaicML aimed to demonstrate that there are chip alternatives available beyond Nvidia, aligning with their strategy to support an open and easily implementable software ecosystem for AI training and inference on AMD hardware.
AMD expressed its support for MosaicML’s findings, stating that they reinforce their commitment to collaborating with the company to optimize their software. As AMD continues to fine-tune its software, customers can explore chip options beyond Nvidia, as highlighted by Mosaic’s research.
Data Table: Comparison of AI Chips – AMD MI250 vs. Nvidia A100
| Chip Model | Performance (Relative to Nvidia A100) |
|---|---|
| AMD MI250 | Approximately 80% |
| Nvidia A100 | 100% |
Note: The above data represents the performance comparison between AMD’s MI250 and Nvidia’s A100 chips. AMD’s chip achieves approximately 80% of Nvidia’s performance.
Disclaimer: The information provided in this article is based on MosaicML’s evaluation and the statements made by its Chief Technology Officer.
