NetEase AI Infra campus hiring

Project experience

1. Describe in detail the main modules and core technical contributions of the AI-related projects on your resume

Inference optimization

2. When designing a high-throughput, low-latency model inference service, what architecture-level and engineering-level issues matter most

C++ basics

3. Explain the traits and differences of stack, heap, and static/global storage in program memory
4. Main differences between new/delete and malloc/free, and why C++ prefers the former
5. Deep copy versus shallow copy, and when you must use deep copy
6. Design intent, use cases, and differences among std::unique_ptr, std::shared_ptr, and std::weak_ptr
7. How the vtable implements runtime polymorphism; semantic and usage differences between virtual and pure virtual functions
8. Memory alignment and compiler struct alignment rules; compute sizeof for a given struct
9. C++11 rvalue references and move semantics, and how they enable efficient data transfer

CUDA programming

10. Basic idea of the SIMT programming model and the hierarchy of thread, block, and grid
11. Where thread-local variables live in a CUDA kernel, and how that relates to register allocation
12. How shared memory cuts repeated global-memory access (tiled matrix multiply as an example)
13. Warp shuffle: the idea and its advantages for in-warp exchange and reduction
14. How vectorized load/store instructions (e.g. float4, int4) enable coalesced access and what performance you gain
15. Causes of shared-memory bank conflict, with examples
16. Avoiding bank conflict via padding or memory-layout changes

Fundamentals

17. How CPU caches work, including temporal locality, spatial locality, and common replacement policies

Coding

18. In a stream of many integers, find the top-K most frequent elements in real time (hash table + min-heap), and analyze time complexity