2 months ago
London, United KingdomStaff+
Responsibilities
- Own and extend the distributed runtime and serving stack for the DX-1 accelerator.
- Define scalable inference architecture across tensor, pipeline, and data parallelism, including collective communication, KV-cache management, offload, and memory-aware scheduling.
- Improve reliability through fault handling, graceful degradation, load balancing, recovery, observability, tracing, and diagnostic tooling.
- Drive pre-silicon and bring-up activities through simulation, emulation, FPGA prototyping, analytical modeling, and root-cause analysis.
- Set technical standards, solve cross-team systems problems, mentor senior talent, and shape platform strategy.
Requirements
- Deep systems software experience with hands-on C/C++ and strong runtime, network, and accelerator fundamentals.
- Demonstrated ownership of complex end-to-end systems problems, ideally involving production distributed inference or serving stacks.
- Experience with distributed inference at cluster scale, including parallelism, collective communication, KV-cache and memory management, and distributed reliability.
- Ability to integrate accelerators with PyTorch, JAX, and serving frameworks without treating framework layers as opaque.
- Experience with whole-stack debugging, timeline tracing, workload replay, and architectural performance analysis.
- Strong judgment regarding speed, cost, and quality trade-offs, plus calm management of emerging risks.
- Excellent communication and cross-functional influence across hardware, compiler, and modeling teams.
- Bachelor’s degree or higher in computer science, electrical engineering, mathematics, or a related field.
- Preferred experience includes dataflow or non-GPU accelerators, ASIC or FPGA bring-up, production observability, HPC cluster design, high-speed networking, distributed systems, or heterogeneous compute platforms.
Benefits
- Competitive salary commensurate with experience, skills, and location.
- Meaningful stock options and employer-contributed retirement plans.
- Annual Living-Local Bonus for residences within 20 minutes of the office.
