According to a report by an AI software firm, Advanced Micro Devices (AMD) is making significant strides in the development of artificial intelligence (AI) chips that are approximately 80% as fast as Nvidia’s offerings, with the potential to achieve comparable performance in the future.
Nvidia currently dominates the market for powerful AI chips, which are instrumental in driving the proliferation of AI services like ChatGPT across the tech industry. The soaring popularity of these services has propelled Nvidia’s market value beyond $1 trillion, leading to a shortage of their chips. Nvidia has acknowledged the issue and is actively working to address the supply constraints.
In light of the chip shortage, tech companies are exploring alternative options, and AMD is emerging as a strong contender. MosaicML, an AI startup recently acquired for $1.3 billion, conducted a test to compare the AI chips from AMD and Nvidia.
The evaluation focused on the AMD MI250 and Nvidia A100 chips, both from the previous generation but still highly sought after. MosaicML’s findings indicate that AMD’s chip can deliver 80% of the performance of Nvidia’s chip, largely due to the release of a new version of AMD software late last year and the introduction of a new version of open-source software called PyTorch, backed by Meta Platforms, in March.
Hanlin Tang, the Chief Technology Officer of MosaicML, expressed confidence that forthcoming software updates from AMD will enable its MI250 chip to match the performance of Nvidia’s A100 chip.
Tang emphasized the crucial role of software in the performance of machine learning chips and praised AMD’s advancements in this area. According to Tang, MosaicML leveraged PyTorch and AMD software to train a large language model without needing to modify their existing code base. As a result, if developers can acquire AMD’s chips at an attractive price point, they can seamlessly transition to using them as viable alternatives to Nvidia chips.
MosaicML specializes in providing software solutions that facilitate the creation of AI systems within a company’s own data centers, eliminating the need to rely on external providers like OpenAI for access to AI systems such as ChatGPT. The research conducted by MosaicML aims to highlight the range of chip options available to its customers beyond Nvidia.
In response to the findings, AMD expressed its commitment to fostering an open and user-friendly software ecosystem for AI training and inference on its hardware. The company acknowledged the significance of Mosaic’s results and confirmed its ongoing collaboration with MosaicML to optimize its software further.