Software Development Engineer - C++, Networking
Overview:
As a PMTS, you will own end-to-end delivery of scale-up networking and GPU software for Instinct GPUs, spanning hardware, firmware, drivers, and ROCm software. You'll lead cross-functional architecture, bring-up, and optimization to deliver high performance and reliability for large-scale AI training and inference. Collaborating with silicon, hardware, and validation teams, you'll drive platform bring-up from pre-silicon to production. This role blends technical leadership with hands-on debugging across multiple layers, shaping AMD's AI infrastructure.
Responsibilities:
- Define and drive architecture for scale-up switch fabrics connecting AMD Instinct GPUs in rack-scale systems, including Ethernet-based fabrics and standards like UALink and Ultra Ethernet
- Lead switch software integration (SONiC, SAI, switch SDKs) with link bring-up, SerDes tuning, FEC/BER characterization, topology discovery, routing, congestion management and telemetry
- Architect and develop GPU firmware components (system management, power/thermal, security, RAS, fabric/link management) and manage firmware packaging and update flows
- Lead development and optimization of Linux kernel drivers for Instinct GPUs (amdgpu/KFD) including memory management, GPU-GPU communication, partitioning, virtualization, error handling and recovery
- Drive ROCm stack enablement and performance on scale-up systems, including runtime, RCCL and integration with PyTorch, JAX and inference engines
- Own new-platform bring-up from pre-silicon through production qualification
- Lead cross-layer debugging and root-cause analysis across silicon, firmware, driver, fabric and application layers
- Define performance, reliability and scalability goals for multi-GPU workloads and drive necessary optimizations
- Collaborate with silicon architects, hardware, switch, cloud/OEM partners and customers to influence roadmaps and technical decisions
- Mentor engineers, set technical direction and engineering best practices across distributed teams
Key requirements:
- Extensive experience in systems software for GPUs, accelerators, NICs or switches
- Strong background in data center networking: Ethernet, high-speed SerDes, FEC, RDMA/RoCE, congestion control, switch ASIC programming
- Familiarity with scale-up interconnects and fabrics (UALink, Ultra Ethernet, Infinity Fabric, NVLink)
- Firmware development experience for SoCs/accelerators (embedded C/C++, boot flows, power management, security, BMC/Redfish/PLDM/MCTP)
- Strong Linux kernel and device driver development (PCIe, DMA, IOMMU, memory management, interrupts)
- Knowledge of GPU software stacks (ROCm/HIP, CUDA) and collective libraries (RCCL/NCCL, MPI)
- Experience with DL frameworks (PyTorch, JAX, TensorFlow) and distributed training/inference at scale
- Proven track record bringing up new hardware platforms and debugging across HW/SW interactions (JTAG, protocol analyzers)
- Proficiency in C/C++ and Python; Go or Rust is a plus
- Experience setting technical direction across teams and excellent communication and leadership skills
- Bachelor's or Master's in CS/CE/EE; PhD a plusexcellent communicationcollaborationleadershipGPU architecture and software stacks (ROCm/HIP)
- Linux kernel and driver developmentPCIe, DMA, IOMMU, memory management