Scaling to 32 GPUs on a Novel Composable System Architecture

Abstract: The development of composable systems architecture marks a significant shift in resource allocation and utilization within data centers. This paper presents a composable architecture scaling up to 32 GPUs on a single node, addressing the technical challenges encountered and the innovative solutions implemented. This design introduces a flexible and dynamic resource distribution mechanism, particularly for GPUs, enabling tailored allocation to meet varying node demands. The architecture's dynamic nature allows for the flexible assignment and reassignment of hardware resources, such as GPUs, to different nodes as required, offering unprecedented capability and flexibility.