Staff Engineer, GPU Systems & Fabric

General Diffusion, Inc.; San Francisco, Northern (CA, KY); Hybrid; USD 180,000 - 250,000; Full time

Job summary

General Diffusion, Inc. seeks a Member of Technical Staff to advance multi-GPU systems and fabric performance.

You will design experiments, profile communication paths, and translate results into actionable models for runtime decisions across the GD-X topology. This role owns fabric behavior between devices, collaborates with Compute World Models, Runtime & Placement, and Measurement & Data teams, and maintains regression benchmarks across hardware configurations and workloads to ensure scalable

Qualifications

• Hands-on experience with multi-GPU or multi-node workloads using collectives (all-reduce, all-gather, reduce-scatter, all-to-all).
• Deep knowledge of GPU communication stacks (NCCL) and PCIe/NVLink behavior.
• Ability to design controlled performance experiments and interpret traces and counters.
• Experience reasoning about topology-aware communication and rank-to-device mapping.
• Experience debugging hangs, timeouts, and throughput issues in Linux distributed environments.

Responsibilities

• Design repeatable experiments measuring collective latency, bandwidth, overlap, and tail behavior.
• Profile end-to-end execution to separate fabric-bound limits from per-device bottlenecks.
• Develop topology-aware configurations and strategies under representative workloads.
• Build telemetry and diagnostic artifacts connecting fabric paths to performance outcomes.
• Maintain regression benchmarks across hardware configurations and software versions.

Skills

Multi-GPU workloads

NCCL / GPU interconnects

Performance experiments

Topology-aware communication

Linux debugging in distributed systems

Tools

NCCL

Profiling tools

Job description

General Diffusion, Inc. seeks a Member of Technical Staff to advance multi-GPU systems and fabric performance.

You will design experiments, profile communication paths, and translate results into actionable models for runtime decisions across the GD-X topology. This role owns fabric behavior between devices, collaborates with Compute World Models, Runtime & Placement, and Measurement & Data teams, and maintains regression benchmarks across hardware configurations and workloads to ensure scalable