A Thermal-First Hybrid ASIC–GPU Architecture with AI-Assisted Dual-Path Latency Management

As computational systems continue to scale in performance and density,thermal stability and long-term reliability have emerged as primaryconstraints, often outweighing marginal gains in raw latency or peakthroughput.This proposal presents a hybrid ASIC–GPU architecture that prioritizesthermal management and operational stability through a dual-pathexecution model, assisted by an AI-based advisory system.The proposed design separates execution into a stable baseline path andan opportunistic fast path, with final arbitration retained by GPU andASIC controllers. AI is employed solely as a monitoring andrecommendation layer, ensuring system robustness while avoidingautonomous control risks.