AI Accelerator Runtime Engineer — Equity Options

OpenAI; San Francisco (CA); On-site; USD 266,000 - 445,000; Full time

Job summary

OpenAI is seeking a Software Engineer for the AI accelerator Runtime in San Francisco. You will build the low-level device runtime that turns compiled programs into efficient execution on our custom accelerator, shaping memory, scheduling, and synchronization.

You will collaborate across compiler, kernel, architecture, verification, and silicon teams to define interfaces, validate behavior, and optimize performance before and after silicon availability.

Qualifications

• Proven experience in low-level systems programming with C/C++ or Rust.
• Designed or built runtimes, drivers, firmware, or OS components.
• Strong understanding of concurrency, synchronization, and memory-ordering.
• Experience with event-based simulators or architecture models is a plus.
• Ability to reason about scheduling, latency, throughput, and resource tradeoffs.

Responsibilities

• Design and implement low-level device runtime for OpenAI silicon.
• Build kernel-launch scheduling, submissions, queues, deps, completion.
• Manage device memories and mappings across workloads.
• Implement synchronization primitives and ordering guarantees.
• Define interfaces between runtime, drivers, firmware, kernels.
• Use cycle-accurate simulators to develop runtime before silicon.
• Diagnose concurrency and memory-ordering issues across HW/SW.
• Create tests, tracing, profiling, and reproducible workloads.
• Collaborate with architecture and silicon teams to improve interfaces.

Skills

C/C++

Rust

Systems programming

Concurrency

Driver development

Memory management

Tools

Runtimes

Drivers

Firmware

OS components

Accelerator software

Job description

OpenAI is seeking a Software Engineer for the AI accelerator Runtime in San Francisco. You will build the low-level device runtime that turns compiled programs into efficient execution on our custom accelerator, shaping memory, scheduling, and synchronization.

You will collaborate across compiler, kernel, architecture, verification, and silicon teams to define interfaces, validate behavior, and optimize performance before and after silicon availability.