Most teams can’t justify a full-time HPC specialist. We are that team — on tap. PeakOps installs, manages and expertly supports high-performance computing, in your data center or on the cloud.
HPC is our core business — not a side feature. We own the hard infrastructure so your researchers and engineers spend their time on the science, not on the scheduler.
We stand up clusters end to end — Slurm, accounting, fabric, storage, GPU stacks, scientific software. On-prem, air-gapped, or on AWS. Built to your workloads, not a template.
Health monitoring, queue and QoS tuning, fair-share, user and job administration, capacity planning. One clear pane of glass over infrastructure that is usually opaque.
Named HPC engineers on an SLA. When a node drains at 2am or a job won’t schedule, you reach people who have run this before — not a first-line ticket queue.
A straightforward path, whether you’re starting from bare metal or inheriting a cluster nobody has fully understood in years.
A short engagement to understand your hardware, workloads and constraints — on-prem, air-gapped or AWS. You get a clear picture of where the gaps and the wasted cycles are.
Green field? We build the cluster — Slurm, accounting, storage, GPU stacks. Existing cluster? We attach our management console to your running Slurm without a reinstall and start reading state.
Ongoing management and named-engineer support on an SLA. Proactive QoS and capacity tuning, incident response, and a console your team can actually read.
The hard part of HPC — Slurm, fabric, GPUs, spend — was never meant to be yours to carry. We fold raw compute into one control plane you can actually read — and run with a Enter single command.
For the cloud, a product you drive yourself. For your data center, a service where our engineers drive for you. Same team, same standard, either way.
A managed service for on-prem HPC — our engineers handle cluster health, Slurm scheduling, users and QoS, capacity planning and incidents. Runs fully offline, backed by an SLA, with a shared console for full visibility.
| NODE | STATE | LOAD |
|---|---|---|
| gpu-a100-01 | alloc | 94% |
| gpu-a100-02 | alloc | 88% |
| cpu-hi-07 | idle | 6% |
| cpu-hi-08 | mix | 51% |
| cpu-hi-09 | drain | 0% |
Cloud HPC orchestration on AWS ParallelCluster. Slurm-aware auto-scaling, per-project budgets, and a Golden-AMI marketplace of ready scientific stacks — live today.
University labs and national research groups that need a fair, well-run cluster — without hiring a dedicated HPC sysadmin per department.
CFD, FEA, crash and CAE teams running MPI jobs at scale. We tune the fabric, the scheduler and the software stack so solvers spend cycles solving.
GPU clusters for training and fine-tuning. Fair-share across teams, honest utilization, and no idle A100s while jobs pile up in the wrong partition.
Defense, public sector and IP-sensitive R&D that can’t touch the public cloud. We run the service fully offline, talking only to your control plane.
HPC done badly is silent waste — idle GPUs, stuck queues, frustrated researchers. Done right, it disappears into the background and simply works. That difference is the whole product.
HPC is the whole company. Slurm internals, fabric tuning, MPI, GPU scheduling, scientific software — the unglamorous expertise most teams can’t justify hiring for full-time. It’s all we do.
We compete on outcomes and response time, not on being the cheapest license. Named engineers who learn your cluster, real SLAs, and a partner who owns the problem alongside you.
On-prem stays on-prem — air-gapped where it must be. In the cloud, everything runs inside your own AWS account with zero credential sharing. Sovereignty is the default, not an upsell.
We run standard Slurm on standard hardware. Our service attaches to what you already have and reads it through Slurm’s own interfaces — end the engagement any time and your cluster keeps running exactly as it did.
If yours isn’t here, ask an engineer directly — we don’t hide behind a form.
Both, and it’s one team. Our Managed On-prem HPC service runs the cluster in your own data center; PeakOps Eleven, our cloud product, orchestrates HPC on your AWS account. Many customers run a hybrid — steady workloads on-prem, bursts to the cloud — and we cover the whole picture.
No. We attach to your existing Slurm through the interfaces it already exposes — scontrol, squeue, sacctmgr and slurmrestd where enabled — and read accounting from slurmdbd. We run alongside your cluster from a separate management host; slurmctld is never replaced.
Yes. For secure sites we deliver everything offline — all images and dependencies vendored, no phone-home, no telemetry, no runtime downloads. We talk only to your cluster’s control plane over your internal network, and work within your access controls. Updates arrive as a signed bundle you stage on your schedule.
Named HPC engineers on an SLA — people who learn your cluster, not a rotating first-line queue. Depending on tier that means business-hours or 24×7 response, proactive QoS and capacity tuning, and on-site or air-gapped support for mission-critical sites.
The on-prem service is a monthly management retainer per cluster (plus a one-time onboarding fee), paired with a support tier scoped to how critical your HPC is — not a software license. Eleven is usage-based on AWS, running in your own account. Research and academic pricing is available. Talk to us and we’ll scope it honestly.
One call. We run a short assessment of your hardware, workloads and constraints, then propose the smallest engagement that solves your actual problem — whether that’s a full build or just taking over day-2 management of a cluster that needs a steadier hand.
Standing up on-prem HPC, wrangling Slurm, or bursting to AWS? Tell us what you’re running. A PeakOps engineer will get back to you — not a bot.
hello@peakops.co