Abstract
For certain traffic types, Field Programmable Gate Arrays (FPGAs) can outperform Graphics Processing Units (GPUs) due to their ability to execute highly parallel and customizable computing tasks efficiently. Recognizing this potential, this submission proposes leveraging Segment Routing traffic engineering (SR-TE) to intelligently direct specific artificial intelligence (AI) workloads to either FPGAs or GPUs based on traffic type, thereby optimizing performance and resource utilization.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Smith, David; Camarillo, Pablo; Abdelsalam, Ahmed; and Filsfils, Clarence, "OPTIMIZING AI WORKLOADS BY OFFLOADING TRAFFIC TO AN FPGA VERSUS A GPU BASED ON TRAFFIC TYPE", Technical Disclosure Commons, (February 05, 2026)
https://www.tdcommons.org/dpubs_series/9277