Server workloads share many of the high-performance compute characteristics, making them a particularly fitting target for SIMT-based processing. At the same time, stringent latency requirements, intensive control-flow, large instruction working sets, lack of memory-level-parallelism, reliance on virtual memory, limited need for floating point units, and frequent interaction with the operating system make server applications unsuitable for existing SIMT designs, especially GPUs. We observe that there exists a large gap between an ideal high-performance energy-efficient SIMT-based server micro-architecture and what conventional CPUs and GPUs can offer. In this project, we aim to remedy this deficiency by setting the groundwork for SIMT-based many-threaded server processors that close this gap.