GPU Supercomputing Engineer for Large-Scale ML Training

Bison; San Francisco (CA); On-site; USD 350,000 - 475,000; Full time

Benefits offered by this job

Health/dental/vision

Unlimited PTO

Parental leave

Relocation support

Job summary

Thinking Machines in San Francisco is seeking an engineer to design, build, and operate the GPU supercomputing environment that powers large-scale training and inference. You will deliver high-performing, reliable compute so researchers can move fast at scale.

You will automate clusters, extend scheduling with Kubernetes or Slurm, monitor metrics, and collaborate with researchers to optimize parallelism and performance.

Qualifications

• Bachelor's degree in computer science, engineering, or similar.
• Proficiency in Python or Rust.
• Experience operating large-scale clusters and container orchestration (Kubernetes or Slurm).
• Ability to own projects end-to-end and work across stacks.

Responsibilities

• Operate and automate large GPU clusters including provisioning and capacity planning.
• Write software that abstracts cluster management for training and inference.
• Extend scheduling/orchestration for topology-aware placement, preemption, quotas, and fair-share multi-tenancy.
• Monitor and improve operational metrics for speed, reliability, and error recovery.
• Build storage paths for datasets, checkpoints, and logs with retention and lineage.
• Collaborate with researchers to unblock scale runs and advise on parallelism and performance trade-offs.

Skills

Python

Rust

Kubernetes

Slurm

Cluster ops

Cross-functional collaboration

Education

Bachelor's degree in CS

Tools

CUDA

NCCL

PyTorch

TensorFlow

Job description

Thinking Machines in San Francisco is seeking an engineer to design, build, and operate the GPU supercomputing environment that powers large-scale training and inference. You will deliver high-performing, reliable compute so researchers can move fast at scale.

You will automate clusters, extend scheduling with Kubernetes or Slurm, monitor metrics, and collaborate with researchers to optimize parallelism and performance.