AMD is setting ambitious goals for its forthcoming MI450 AI GPU, asserting it will deliver top-tier performance across various AI tasks, including training and inference. Forrest Norrod, AMD’s Executive VP & GM of Data Center Solutions, recently revealed details at the Goldman Sachs Communicopia+ Technology Conference, emphasizing the transformative potential of MI450.

“We will systematically build out training capability in MI355, both on the silicon as well as the software side.”

AMD’s strategy aims to secure its leadership in the GPU domain, targeting key areas such as distributed inference and reinforcement learning. Norrod hopes MI450 will parallel the success seen with AMD’s EPYC CPUs, marking a pivotal moment for the company in the GPU sector.

With aspirations to amalgamate its AI and gaming GPU architectures into a unified system named UDNA, AMD is looking to enhance its standing in the graphics market. Norrod expressed that gaining a meaningful share in the market is crucial, aiming for around 20% as a significant milestone.

The MI450, while not intended for gaming, is positioned as a cornerstone of AMD’s strategy to compete directly with Nvidia’s technologies in the evolving landscape of AI applications.