LENZO Chief Architect and NAIST Professor Yasuhiko Nakashima has published a new article exploring low-power parallel computing using dataflow architectures and their potential to improve performance and energy efficiency for modern workloads.
The paper compares traditional processor designs - including CPUs, GPUs, and "manycore" architectures - with dataflow-based execution models, highlighting how pipeline driven computation can significantly reduce memory bottlenecks and improve power efficiency.
The article also introduces the CGLA architecture, a dataflow computing approach developed through years of research at NAIST and now being commercialized by LENZO. By enabling long-running pipeline execution with predictable memory behavior, CGLA aims to deliver highly efficient computation for emerging workloads such as AI and large-scale numerical processing.