Principal Scientist - Software/Hardware Co-design
Core
Build accurate AI performance models and define algorithm-hardware co-design features for next-generation chips to optimize compute efficiency and throughput.
Role type
Principal Scientist (AI/Hardware Co-design)
Builds
Next-generation GPU architecture and AI acceleration platforms
Domain
AI hardware, high-performance computing, chip architecture
Deliverable
production ML models
Required skills
AI performance modeling, algorithm-hardware co-design, large language model principles, hybrid parallelism, low precision data formats, sparsity, P/D splitting, microarchitecture understanding, quantitative analysis of compute/memory/IO/interconnect
Preferred skills
Low-level computing algorithm development, AI accelerator design, large scale parallel computing, high performance computing system design, familiarity with AI chip microarchitecture
Technologies
AI software stack (operators, compilers, acceleration libraries, frameworks), mainstream large model training and inference algorithms
Responsibilities
Build accurate and universal AI performance models based on mainstream AI acceleration technologies; Track emerging hardware designs and conduct survey analysis to identify key cutting-edge technologies; Cooperate with AI research teams to identify performance bottlenecks and define co-design features; Perform performance modeling of AI workloads on different hardware specs for quantitative analysis; Lead the team in achieving breakthroughs in acceleration algorithms balancing model quality and compute efficiency.