Job Description:
Lead the design and development of a high-compatibility CUDA translation layer(software-hardware co-design) to enable seamless migration of existing CUDA-basedLLM workloads to our chip. Define the compatibility scope (APIs, kernels, librariessuch as cuBLAS/cuDNN) and optimize translation efficiency to minimize performanceoverhead,Collaborate with the GPU core team to align hardware features withcompatibility requirements, and work with customer engineering teams to validatecompatibility with internet giants’ production workloads.
任职要求
Requirements:
8+ years of experience in GPU software-stack development, with deep knowledge ofCUDA runtime, driver, and compiler (NVCC).Proven experience in building APl compatibility layers or cross-platform GPUsoftware tools.Familiarity with LLM-inference frameworks (e.g., TensorRT, ONNX Runtime) andCUDA-accelerated libraries.
Strong problem-solving skills for debugging complex compatibility and performanceissues.
