Hardware Engineer
onsite
· 难度 5/5
Recruiter screen,然后技术电话面考 Verilog 和计算机体系结构。全天 onsite:RTL 设计题、post-silicon debug 案例分析、和多位 architect 的行为面,以及深挖你参与过的一次 tape-out。通过的人很少,他们在架构岗位上招人非常保守。
由 AI 翻译,查看原文
- Tell me about a critical bug you found late in silicon validation. How did you triage?(讲讲一次你在 silicon validation 后期发现关键 bug 的经历。你怎么做分流和优先级处理?)
- Describe disagreeing with a senior architect on a design tradeoff. How did you resolve it?(描述一次你和 senior architect 在设计取舍上意见不一致。你们怎么解决的?)
- Walk me through your debug methodology on a timing-closure failure.(说说你在 timing-closure 失败时的 debug 方法论。)
- How do you balance feature scope vs. tape-out timeline?(你如何平衡功能范围和 tape-out 时间线?)
Software Engineer
virtual
· 难度 5/5
先 recruiter screen,然后技术电话面:一道 CUDA kernel 优化题。线上 onsite 共五轮:两轮 coding(一轮 CPU,一轮 GPU)、system design(分布式训练 pipeline)、对过往项目深挖,以及行为面。系统基础要求非常高。
由 AI 翻译,查看原文
- Tell me about a time you optimized code by 10x or more. What was the bottleneck?(讲讲一次你把代码优化到 10 倍或更高的经历。瓶颈是什么?)
- Describe a project where you had to make hardware/software tradeoffs.(描述一个你必须在软硬件之间做权衡的项目。)
- Walk me through how you'd debug a kernel that's slower than expected.(说说如果一个 kernel 比预期更慢,你会怎么 debug。)
- How do you stay current with the ML accelerator landscape?(你如何跟进 ML 加速器领域的最新进展?)
Research Scientist
virtual
· 难度 5/5
NVIDIA Research 的 research 岗位。先电话面试,然后做 research presentation。线上 onsite 共 6 轮:和多个团队的 research talk、论文技术深挖、coding(实现 ML 算法),以及行为面。整体非常偏学术风格,重点看研究潜力。
由 AI 翻译,查看原文
- Walk me through your most impactful research contribution.(带我过一遍你最有影响力的研究贡献。)
- How do you identify promising research directions?(你如何识别有前景的研究方向?)
- Describe a time you had to pivot your research approach due to unexpected results.(描述一次因为意外结果,你不得不调整研究方向/方法的经历。)
- How do you balance publishing with product impact?(你如何在发表论文和产品影响之间做平衡?)