Staff Software Engineer - GPU System Software, Onboard Execution

Zoox•Boston, MA

About The Position

Zoox is seeking a technical leader for onboard GPU execution within its Software Performance organization, which owns end-to-end performance of the onboard software stack, including profiling and observability, GPU compute, and the compute platform. The role defines how AI workloads run on Zoox vehicles and sets the strategy for delivering more capability with less onboard compute. The engineer will establish an end-to-end technical vision for scheduling, sharing, isolating, and sizing perception, prediction, and planning workloads across onboard GPUs. They will work closely with Perception, Prediction, Planning, ML Infrastructure, and Hardware teams, and serve as a senior technical voice for onboard compute across the company. The work involves balancing performance, cost, power, safety, and real-time requirements across current and future vehicle platforms.

Requirements

Bachelor's degree in computer science, computer engineering, or a related field.
10 or more years of experience in systems or performance engineering, including leading GPU or accelerator architecture across teams, or building and leading a performance engineering team.
Deep expertise in the CUDA platform, including its programming model, runtime, and driver.
Expertise in NVIDIA GPU microarchitecture, including Ampere through Blackwell, Orin, and Thor.
Proven record of setting technical direction and delivering large, measurable end-to-end improvements in latency, utilization, or cost.
Working knowledge of ML inference optimization, including TensorRT, XLA, Triton, quantization, and fusion.
Expert C++ skills on Linux.
Ability to align ML, systems, hardware, and leadership stakeholders on a shared technical plan.

Nice To Haves

Real-time or embedded GPU experience in autonomy, robotics, AR/VR, games, or HPC.
Production GPU multi-tenancy experience, including MPS, MIG, or priority streams.
Experience with new GPU architecture or platform bring-up, benchmarking methodology, workload modeling, or performance projections.

Responsibilities

Architect onboard GPU execution by defining how perception, prediction, and planning workloads share GPUs, including scheduling, multi-tenancy, isolation, and deterministic latency under contention.
Own a multi-year roadmap to reduce onboard compute per vehicle by setting latency and utilization budgets and delivering improvements across models, runtimes, middleware, and hardware.
Guide next-generation compute decisions, including GPU and SoC selection, topology, and memory budgets, using workload models and performance projections.
Lead performance bring-up of each new compute platform.
Set direction for the inference stack, including TensorRT, CUDA, custom kernels, and quantization, and partner with ML teams to design models for their target hardware.
Lead and mentor senior engineers, align directors and VPs through design documents, and make decisions when performance, safety, and schedule conflict.