NPU Performance Optimization Engineer(AI Processor)

Job Description:

Enable AI algorithms, models, and applications on AI-related intellectual property products.

Complete theoretical performance analysis of models, and based on profiling data, quantitatively analyze the primary performance bottlenecks of models on specific AI processors, and provide corresponding optimization strategies.

Lead model performance delivery, independently optimize several critical bottleneck points, and collaborate with internal software/hardware teams to push model performance optimization close to hardware limits.

Summarize major software and hardware-level performance issues as input for product improvements.

Required Skills and Experience:

Solid programming skills in C/C++ and Python, familiar with open-source development tools.

Proficient in AI and deep learning models, algorithms, workload computation, and bandwidth analysis.

Familiar with low-level algorithm implementation, software stacks, and hardware pipelines.

Familiar with common AI algorithm acceleration methods such as quantization, pruning, and sparsification, with the ability to explore cutting-edge algorithms.

Actual implementation experience in CUDA/OpenCL performance tuning is preferred.