AI Software System Designer(GPU Kernel)

加入AMD，你将参与设计并优化GPU Kernel，为AI训练与推理提供高性能软件解决方案

KEY RESPONSIBILITIES:

PREFERRED EXPERIENCE: Ability to work independently, define project goals and scope, and lead your own development effort. Solid communication skills, in both English and mandarin. Excellent in GPU kernel primitive like Attention (FA, PA, MLA, linear Attn etc.), MOE, TOPK design (algorithms) and development (with Triton/DSL, Cutlass/CK, CUDA/HIP, PTX/ASM etc.). Excellent programming skills in Python, C++ and software skills, including debugging and performance analysis. Experiences with model inference optimization process like GEMM/convolution tuning, graph optimization and operator fusion. Experiences with AI frameworks(e.g. vLLM, Sglang, Megatron-LM, Deepspeed, TensorRT, TensorRT-LLM etc.). Knowledge of compiler (Torch, Triton, LLVM, XLA HLO, graph) is a plus. Knowledge of Linux ROCm/CUDA runtime and KMD/UMD driver is a plus. Knowledge of AI distribution solutions (i.e. EP/SP/CP/TP/PP/DP, DeepEp, DualPipe, PD aggregation etc., KV cache transfer and storage). Knowledge of AI distributed network communication with multi-GPU and multi-node collective communication primitives (NCCL/RCCL), NIC/GPU drivers for RDMA/GDR and high-speed network etc. Knowledge of Linux OS/driver, CI and toolchain (profiler/DCGM) development and debugging.