AMD Forward-Deployed Engineer Interview Questions
The questions to prepare for a AMD Forward-Deployed Engineer interview. Questions from real interview reports rank first. Updated daily.
Select a GPU matrix multiplication tile by minimizing coalesced global transactions while respecting shared-memory usage.
AMDSimulate Ring-AllReduce reduction and gathering across GPU nodes while reporting communication rounds and point-to-point messages.
AMDTests ability to apply GPU memory hierarchy techniques to improve convolution performance.
AMDExplains KV-cache memory and decode-time reuse, then shows how PagedAttention reduces fragmentation for AMD GPU inference.
AMDDiagnose low AMD GPU utilization by separating input, kernel, CPU, synchronization, and RCCL communication bottlenecks.
AMDTests how PyTorch 2.0 torch.compile captures, specializes, fuses, and executes graphs differently from eager GPU execution.
AMDSign up to see every question
Create a free account to unlock this list and practice real interview questions.
Optimize a PyTorch input pipeline so AMD GPU training stays compute-bound instead of waiting on CPU data loading.
AMDTests ability to design scalable distributed training strategies for very large models on AMD hardware.
AMD