GPU Cluster Engineer - Systems & Platform
$150k - $220k/yr+ Equity
Senior
Bachelor's, Master's
San Francisco, CA, USA
In Person
DevOps & Infrastructure
About the job
Requirements
- At least 5 years of systems or infrastructure engineering experience with significant GPU cluster, high-performance computing, or large-scale machine learning infrastructure experience.
- A Bachelor's or Master's degree in Computer Science, Computer Engineering, Electrical Engineering, or a related technical field.
- Deep expertise in Linux internals, including kernel modules and DKMS, systemd, cgroups, NUMA, and system performance tuning.
- Hands-on experience with NVIDIA CUDA and/or AMD ROCm driver and runtime stacks on modern accelerators, including kernel-level debugging.
- Production Kubernetes experience with GPU workloads and working knowledge of HPC schedulers such as Slurm or Run:AI, or deep Slurm experience and working Kubernetes knowledge.
- Strong configuration management experience with Ansible or SaltStack and Git-based, code-reviewed infrastructure workflows.
- Experience with provisioning and image tools such as Packer, MaaS, Foreman, or Terraform for automated, reproducible node builds.
- Client-side experience with distributed filesystems such as Lustre, GPFS, or Weka, and checkpoint I/O optimization.
- Fluency with Docker or containerd and the NVIDIA Container Toolkit or ROCm equivalent.
- Proficiency in Python and Bash for automation and tooling.
- Working knowledge of NCCL and RDMA networking, including InfiniBand or RoCE and GPUDirect, and PyTorch or JAX runtime behavior.
Responsibilities
- Own the node software definition, including versioned operating system images, kernel tuning, GPU and network interface driver stacks, and the automated pipeline that takes a node from base operating system to production-ready.
- Build automated acceptance suites using DCGM diagnostics, NCCL or RCCL tests, bandwidth and topology checks, and HPL to validate and gate nodes before they enter a scheduler pool.
- Execute rolling kernel, driver, and toolkit upgrades with minimal disruption to running workloads; enforce configuration consistency, detect drift, and maintain the driver, CUDA or ROCm, and framework compatibility matrix across the fleet.
- Automate detection of unhealthy nodes, including Xid or ECC errors, link flaps, and thermal throttling, and implement cordon, drain, reboot, and re-image workflows with handoff to Hardware Operations for physical repair or RMA.
- Manage node and cluster configuration through Ansible or SaltStack playbooks in Git, with peer-reviewed changes, CI validation, and canary rollouts before fleet-wide deployment.
- Build and maintain image and provisioning tooling, such as PXE, MaaS, or Packer, so new or re-imaged nodes are reproducible rather than hand-crafted.
- Develop Python and Bash tooling for cluster operations, health reporting, and workflow automation.
- Deploy and operate GPU-enabled Kubernetes for inference workloads, including NVIDIA GPU Operator, device plugins, node feature discovery, topology-aware scheduling, and MIG or MPS partitioning where appropriate.
- Operate Slurm or Run:AI for multi-node training, including partitions, QoS, preemption, accounting, and container integration using enroot or pyxis.
- Maintain base container images and registries, and the NVIDIA Container Toolkit or ROCm container stack; keep training and serving images lean, current, and reproducible.
- Build, deploy, and debug the full accelerator stack for NVIDIA and AMD, including CUDA toolkit, cuDNN, NCCL, Fabric Manager, ROCm, RCCL, kernel modules using DKMS, GPUDirect RDMA or Storage, and the MOFED or DOCA RDMA software stack.
- Maintain curated, optimized PyTorch and JAX environments with managed dependencies and versions for researchers and production services.
- Tune NCCL or RCCL across NVLink or NVSwitch and InfiniBand or RoCE fabrics, ensure topology-aware job placement, and run continuous communication and throughput benchmarks to detect regressions.
- Own escalation for NCCL hangs and timeouts, CUDA memory leaks, ROCm kernel crashes, straggler nodes, and unexplained throughput drops.
- Own software-layer monitoring using DCGM exporter, Prometheus and Grafana, and alerting, as well as job-level GPU utilization and cluster efficiency reporting.
Desired Qualifications
- Experience directly supporting foundation model training teams, including multi-node job failure debugging, checkpoint pipeline tuning, and framework-level performance triage, ideally in a startup or research-heavy environment.
- Experience deploying and tuning inference or serving stacks such as vLLM, Triton Inference Server, or TensorRT-LLM for latency and throughput targets.
- Experience with GPU or system profiling tools such as Nsight Systems, Nsight Compute, rocprof, perf, or eBPF.
About the company
Sciforium provides an AI infrastructure stack with a serverless inference platform, giving access to a mix of open-source and proprietary models through a single API. It runs on its own custom-optimized AMD hardware, delivering models without shared cloud servers to reduce costs, boost performance, and improve data privacy. The company differentiates itself by owning the full stack—hardware, software, and models—and by pursuing byte-native multimodal foundation models. Its goal is to simplify and speed up production-ready AI deployment while lowering costs and complexity for large-scale models.
Company Size
11-50
Company Stage
N/A
Total Funding
N/A
Headquarters
San Francisco, California
Founded
2024
Get referred to Sciforium
See people who can refer or advise you
Simplify's Take
What believers are saying
- SignalFire still lists Sciforium current as of June 2026.
- LinkedIn shows aggressive hiring for GPUs, ML, and growth in August 2026.
- First tranche of GPU capacity targets October 2026, suggesting near-term productization.
What critics are saying
- October 2026 capacity depends on converting one industrial facility on schedule.
- Twelve jobs across GPU, data center, and model-serving roles signal heavy burn.
- No public customers or revenue announcements expose Sciforium to existential platform credibility risk.
What makes Sciforium unique
- AMD-backed vertical stack ties hardware, serving, and models into one platform.
- Byte-native multimodal foundation models and inference platform share the same infrastructure.
- Custom AMD GPU operations target lower TCO than shared-cloud inference competitors.
Help us improve and share your feedback! Did you find this helpful?
Benefits
- Health Insurance
- Dental Insurance
- Vision Insurance
- 401(k) Retirement Plan
- Meal Benefits
- Company Equity