XPENG is a leading smart technology company at the forefront of innovation, integrating advanced AI and autonomous driving technologies into its vehicles, including electric vehicles (EVs), electric vertical take-off and landing (eVTOL) aircraft, and robotics. With a strong focus on intelligent mobility, XPENG is dedicated to reshaping the future of transportation through cutting-edge R&D in AI, machine learning, and smart connectivity.

Our pioneering first-generation NPU, utilizing DSA architecture, has successfully entered mass production. We're currently validating the architecture of our second generation and are making the strategic decision to transition towards General Purpose GPU (GPGPU) architecture.

We're completely overhauling our software stack and embracing the CUDA ecosystem. Our goal is to achieve over 90% compatibility with cuBLAS/cuDNN on Linux across PCIe and CXL connections, scaling from single-GPU setups to 2-GPU chiplet configurations, all while delivering at least 1.3 times the performance of existing solutions on Transformer and Stable-Diffusion workloads.

Job Responsibilities:

Software Technical Strategy

Develop and refine a comprehensive 3-year roadmap for a software stack compatible with CUDA, encompassing Runtime, Driver, Compiler, Profiler, Debugger, and AI acceleration libraries Define binding specifications that link our upcoming GPU ISA to CUDA APIs, ensuring forward compatibility with CUDA 12.x features Evaluate and integrate the latest technological advancements: CUDA Graph, Transformer Engine, virtual memory management, CUDA dynamic CUTLASS 3.x, TMA, Blackwell FP4, among others Architecture & Design Create a modular, layered Runtime architecture: CUDA → HAL → Kernel → Hardware, applicable across emulators, FPGA prototypes, and actual silicon Define the task launch protocol, including Queue, Stream, Event, and Graph, as well as the memory model Design a dual-mode (JIT & offline) compiler supporting LTO, PGO, Auto-Tuning, and efficient PTX→ISA microcode caching Develop GPU virtualization schemes(MIG) that work across processes and containers

Performance & Observability Implement an end-to-end performance model: Python API → CUDA Runtime → Driver → ISA → Micro-architecture → Board-level interconnect Build an observability platform: Nsys-compatible traces, real-time Metric-QPS dashboards, and an AI Advisor for identifying bottlenecks automatically Manage internal AI benchmarks as the single source of truth. Benchmark includes MLPerf Inference, Stable Diffusion XL, and 70B LLM Cross-functional Collaboration Co-design ISA which compatible with CUDA Compute Capability 12.x with our hardware architecture team Collaborate with AI framework teams (PyTorch, TensorFlow, JAX, ONNX Runtime) to build fully reusable kernel libraries Partner with Cloud and K8s teams to co-develop Device Plugins, GPU Operators, and RDMA Network Policies Minimum Requirements: 10 years + in systems software, with at least 5 years in designing CUDA Compute stacks Led end-to-end development of a GPU Runtime or AI acceleration library generation Comprehensive mastery of PTX/SASS, CUDA Driver API, and cuBLAS/cuDNN/cuFFT internals; experience with LLVM NVPTX backend Profound understanding of GPU micro-architecture, including SM architecture, Warp Scheduler, Shared-Memory conflicts, and Tensor Core pipelines Proficiency with PCIe/CXL/RDMA topologies, NUMA settings, and GPU Direct RDMA/Storage

The base salary range for this full-time position is $241,800 - $409,200 in addition to bonus, equity and benefits. Our salary ranges are determined by role, level, and location. The range displayed on each job posting reflects the minimum and maximum target for new hire salaries for the position across all US locations. Within the range, individual pay is determined by work location and additional factors, including job-related skills, experience, and relevant education or training.

We are an Equal Opportunity Employer. It is our policy to provide equal employment opportunities to all qualified persons without regard to race, age, color, sex, sexual orientation, religion, national origin, disability, veteran status or marital status or any other prescribed category set forth in federal or state regulations.