GPU Chief Architect
2 days ago
Cambridge
We are seeking a highly experienced GPU architect to lead the definition and execution of next-generation mobile GPU architecture, while driving architectural convergence between GPU and NPU toward a coherent xPU sub-system design. This role requires deep expertise in GPU microarchitecture, strong system-level architectural capability, including both hardware and software, and a thorough understanding in graphics and AI common workload. A proven track record of delivering related sub-system IP or complex SoC silicon is highly desirable. The successful candidate will lead the effort in shaping a converged xPU architecture native for future AI compute, optimised for performance, power efficiency, and silicon area in the next generation mobile compute platforms. Key Responsibilities: xPU Converged Architecture Design • Based on 1st order principle, analyse and characterise future mobile graphics and AI workload, redefine an xPU (GPU & NPU) converged architecture, including hardware and software, from the ground up that is optimal for future applications., • Ensure compatibility or easy transition from the old architecture., • Define unified or partially unified execution resources (vector, scalar, tensor units), • Develop shared scheduling and workload dispatch mechanisms for graphics and AI, • Design resource sharing and isolation strategies under mixed workloads, • Evaluate architectural trade-offs between dedicated and converged compute blocks, • Mobile GPU Architecture Leadership, • Ensure the timely delivery of next-generation mobile GPU architecture and long-term roadmap, • Lead evolution of shader cores, execution pipelines, and cache hierarchy, • Drive performance, power efficiency (Perf/W), and area efficiency (Perf/mm²), • Provide architectural leadership from concept phase through tape-out, • Memory & Interconnect Architecture, • Define a memory hierarchy strategy for converged GPU/NPU workloads, • The architect shared cache structures and bandwidth arbitration policies, • Optimise on-chip interconnect for heterogeneous compute traffic, • Reduce data movement overhead across compute domains, • System-Level Architecture Collaboration, • Collaborate with CPU, AI software, runtime, and system architecture teams, • Participate in SoC-level power, thermal, and floorplanning trade-offs, • Align hardware architecture with graphics APIs and AI frameworks, • Support performance modelling, workload characterisation, and silicon bring-up Required: • 15+ years of experience in GPU, AI accelerator, or heterogeneous compute architecture, • Deep understanding of GPU microarchitecture (SIMD/SIMT, scheduling, memory systems), • Strong knowledge of tensor/matrix computation and AI acceleration techniques, • Proven experience delivering high-volume silicon, • Expertise in performance modelling and power analysis, • Strong cross-functional communication and leadership capability