With the growing demand for high-performance Artificial Intelligence (AI) applications, the design of efficient server architectures has become critical. Modern AI workloads—such as training deep learning models and processing large-scale data—require significant computational resources, which can be effectively met using distributed computing environments composed of high-performance workstations interconnected by fast networks. To ensure usability, such distributed systems must provide a single-system image, abstracting the complexity of the underlying hardware from end users. In building AI applications for these environments, threads offer a more efficient and practical alternative to traditional processes. Threads have lower overhead for creation, context switching, and communication, making them ideal for implementing fine-grained parallelism in multi-core and multi-processor systems. By utilizing threads, AI servers can efficiently distribute computation across multiple processors, improving throughput and responsiveness. This paper presents a prototype AI server design and investigates how multithreading enhances the performance of AI applications running on parallel processing architectures.
Journal of Theoretical and Applied Information Technology (Mon,) studied this question.