UC San Diego

As we approach the end of Moore's law, the research focus has shifted towards developing hardware accelerator based designs to achieve higher performance with lower power requirements. The key objective while developing a server architecture is to minimize the Total Cost of Ownership (TCO) while improving both performance and energy efficiency. Thus, the trend in such planetary scale applications has been to move from CPUs, GPUs and FPGAs towards arrays of application specific accelerators, known as ASIC Clouds. In this work we present ArCuS, a hardware infrastructure for developing Architectures for ASIC Cloud based Servers. We layout the architectural specifications, RTL description and physical implementation details for the ASIC which aims to serve as a reference and enable rapid prototyping. We design and characterize the proposed architecture across technology nodes and evaluate various metrics for accelerating bitcoin mining application. Finally, we present the cost-performance trade-offs necessary for determining a cost-optimal ASIC Cloud configuration.