- AMD unveils Instinct MI350 Series GPUs, promising a 4x AI compute increase and 35x improvement in inferencing.
- The new MI355X offers a 40% better tokens-per-dollar efficiency compared to competitors.
- Significant industry partnerships with Meta, OpenAI, Microsoft, Oracle, and more.
AMD (AMD, Financial) has announced a significant expansion in its AI capabilities at the Advancing AI 2025 event, launching new hardware, software, and systems aimed at enhancing AI performance. A centerpiece of this unveiling is the Instinct MI350 Series GPUs, which provide a fourfold increase in AI compute power and a 35x improvement in inference capabilities over previous generations. The MI355X, a standout model in this series, offers a 40% better tokens-per-dollar ratio, presenting a competitive edge in the market.
The company also showcased its open rack-scale AI infrastructure, integrating the Instinct MI350 Series accelerators, 5th Gen EPYC processors, and Pensando Pollara NICs. This strategic move is aimed at creating a comprehensive, industry-standard AI platform. AMD's upcoming "Helios" rack, featuring the next-gen Instinct MI400 Series GPUs, is projected to boost performance by ten times, particularly beneficial for Mixture of Experts models.
In software development, AMD's ROCm 7 software stack introduces enhancements that address previous critiques and bolster support for high-performance computing tasks. Notably, AMD has surpassed its five-year goal of improving energy efficiency, achieving a 38x enhancement. Looking forward, AMD has set a 2030 target for further increasing rack-scale efficiency by 20 times, emphasizing the importance of sustainability as AI technologies advance.
The announcement also highlighted partnerships with leading tech companies such as Meta, OpenAI, Microsoft, and Oracle, all utilizing AMD's Instinct accelerators for their AI workloads. These collaborations underscore AMD's expanding influence in the AI sector, with Oracle Cloud Infrastructure planning to incorporate AMD's Instinct MI355X GPUs into its operations, facilitating the development of large-scale AI clusters.
