Both AMD and Google have announced the release of their AI accelerators. AMD has launched the Instinct MI300, a GPU accelerator, while Google has introduced the TPU v5e, a tensor processing unit designed to power AI in the Google Cloud.
AMD is aiming to catch up to Nvidia, a leader in the AI processing field. AMD’s Instinct MI300 chips offer faster AI operations and can be combined to further enhance performance. The company has already gained support from customers and partners such as Dell, HPE, Microsoft, and Oracle, who are either running the chips in their products and services, testing them, or planning to use them in the near future.
The 153 billion transistors and the claimed 17TB/second bandwidth of AMD’s MI300 accelerator offer enterprise IT buyers significant benefits. According to AMD CEO Lisa Su, the accelerator enables faster AI operations, making it well-suited for demanding tasks such as training large language models and answering millions of queries from users worldwide.
Daniel Newman, the founder of Futurum Research, highlighted the importance of the hardware. While performance is crucial, open-source platforms that allow developers to build software and connect it with the hardware are also essential. AMD’s entry into the AI market with open-source capabilities and competitive products challenges Nvidia’s dominance and emphasizes the significance of open-source collaborative ecosystems.
Gartner analyst Chirag Dekate emphasized that many companies still rely on their own GPUs in their data centers, even in the cloud-first era. They may adopt the AMD GPU accelerators to maintain data privacy or protect intellectual property. Dekate also predicted that the combination of hardware, software, and partnerships offered by AMD would help enterprise customers establish their AI operations more quickly.
Google, alongside its general AI model release and plans to integrate generative AI into smartphones, has unveiled the TPU v5e. TPUs power Google’s own AI in various apps and are now available on the Google Cloud Platform. Dekate noted that in the future, enterprise cloud services buyers may choose different AI services powered by chips from different manufacturers depending on the specific requirements of their applications and operations.
Competition is crucial in the AI hardware race to ensure the advancement and viability of AI chips. Don Fluckinger of TechTarget Editorial emphasizes the need for a highly competitive marketplace for AI infrastructure, chipsets, and software to support the transformative potential of generative AI.