Principal System Software Architect, AI/GPU Platforms
Core
Define end-to-end system software architecture for AMD Instinct™ MI400-class GPU platforms, spanning memory management, scheduling, virtualization, and scale-up/scale-out interconnects to enable maximum performance for AI and HPC workloads.
Role type
Principal System Software Architect (AI/GPU Platforms)
Builds
Rack-scale AI training and inference systems using AMD Instinct GPUs
Domain
Semiconductor / High-Performance Computing / AI Infrastructure
Deliverable
production ML models
Required skills
Linux Memory Management, Heterogeneous Memory Management (HMM), GPU/DRM driver development, Cache coherence protocols, GPU Networking (NVLink, UALink, RDMA), Collective communication (RCCL, NCCL, MPI), Data-center fabrics (Infinity Fabric, InfiniBand), ROCm stack, CUDA, Deep learning frameworks (PyTorch, TensorFlow), GPU compute kernels (HIP, Triton), OpenCL, Vulkan, Virtualization, Confidential computing
Preferred skills
Pre-silicon hardware/software interface definition, Multi-GPU topology strategy, Performance bottleneck analysis, Architecture specification creation, Technical engagement with hyperscale customers
Technologies
amdgpu, KFD, ROCr, HSA, ROCm, HIP, CUDA, Triton, PyTorch, TensorFlow, ONNX, MLIR, Infinity Fabric, UALink, RoCE, InfiniBand
Responsibilities
Own end-to-end architecture for GPU memory management, scheduling, RAS, and virtualization; Drive architecture across kernel-mode drivers, user-mode runtimes, and firmware; Partner with silicon architects on pre-silicon definitions; Define software strategy for multi-GPU topologies and interconnects; Establish architecture for reliability and serviceability at scale; Set direction on performance optimization; Produce architecture specifications and reference designs; Act as technical anchor for strategic customers
Seniority
Principal, hands-on IC
