Managed scientific HPC in your own AWS — CADD, cryo-EM & AI on one queue

Your computational research stack, operated in your AWS.

Managed Slurm and the validated open-source scientific stack — CADD, cryo-EM, multi-omics, and AI, on one queue — running in your own AWS account. Your data stays in your VPC; your AWS & NVIDIA credits pay the compute.

3 design-partner slots · BYOC plans from $99/mo · free 30-day pilot

Drug discovery / CADD Cryo-EM & cryo-ET Genomics & proteomics BYOC — your data stays in your VPC
Why this, not the obvious alternatives

Your best people weren't hired to babysit a cluster.

Right now they lose weeks to undifferentiated heavy lifting — building containers, tuning GPUs, nursing spot interruptions, wiring Slurm. Clusterra hands all of it to a validated, managed cluster in your own AWS, so your scientists and engineers spend their time on the science and product only they can build. Two questions usually follow — answered below.

vs. raw AWS

AWS sells the parts. We hand you the working instrument.

Batch, PCS, and ParallelCluster are kits — someone still has to design the cluster, build the containers, wire Slurm, and tune the GPUs. That’s a dedicated HPC/DevOps headcount. We deliver it built, validated, and running on day one.

Spot GPUs at ~60% off — with checkpoint-resume so long jobs survive reclaim.

vs. a SaaS platform

Your AWS & NVIDIA credits pay the compute.

Most biotechs hold $100K+ AWS Activate and $100K+ NVIDIA Inception credits. Because Clusterra runs in your account, every run draws them down directly — so for months, world-class compute costs you effectively nothing.

SaaS bio-platform Bills you directly. Credits can’t pay a third-party invoice — they sit unused and expire while you pay cash.
Clusterra — in your own account Each run draws the credits down. No markup on compute, ever.
who builds it

Built by ex-AWS HPC — the people who shipped the parts.

Clusterra comes out of AWS HPC — built by people who helped shape AWS’s own HPC services. When we say the raw parts aren’t the instrument, it’s from the inside. Built on open-source Slurm — no lock-in.

AWS BatchAWS PCSParallelCluster

Your HPC team for a fraction of one hire — not three (HPC + AI/ML infra + DevOps).

Pick your science

One platform. Three front doors.

The same managed Slurm cluster runs all three. Start with the pain blocking you now — each has a dedicated page with benchmarks, templates, and case studies.

/cryo-et

Cryo-EM & cryo-ET — SPA, STA, tomography

The cluster cryoSPARC and RELION-5 actually need.

For structural-biology teams running RELION-5 / AreTomo / Warp·M / IsoNet. Multi-TB tilt-series stay in your VPC. No academic-cluster waitlist, no $90/TB egress.

  • HIV-1 Gag STA: 3.99 Å for ~$2.30 spot
  • Gold-standard FSC, end-to-end in your account
Open the cryo-EM page →
/multi-omics

Multi-omics — genomics & proteomics

nf-core, AlphaDIA, MAPPs — at cohort scale.

For teams running Nextflow/nf-core, mass-spec, and single-cell pipelines. Cohorts run in parallel, not serial — autoscaling on spot, back to zero when idle.

  • AlphaDIA: $0.05/sample, 13 min wall-clock
  • Sarek + Parabricks 30× WGS: $2.59/sample
Open the multi-omics page →
See it run · a CADD campaign

One command runs the whole pipeline — on your own AWS.

A single sbatch chains structure → dock → co-fold → MD across spot GPUs in your account — 73 minutes, $1.16, every step costed and reproducible. One pass of a longer optimization loop, not the finish line. (Each branch has its own end-to-end demo.)

 ·  EGFR ATP site (7TU3), Enamine Diverse 1K — one sbatch, unedited wall-clock.

The open-source scientific stack your team already cites

· · · · · · · · · · · ·
Time to a working cluster

Cluster live today. First campaign in 30 days.

AWS PCS, DIY4 – 8 weeks
HPC engineer hire + ramp12 – 24 weeks
Clusterratoday
What you get

A campaign workspace, not just a cluster.

A managed Slurm cluster in your AWS, your domain stack operated end-to-end, and every run's history accreting as your program's computational book-of-record.

01 / Campaign workspace

Every run builds your program's provenance record.

Results, parameters, and per-job cost stamps accumulate in your own EFS — a cost-stamped record that lives in your account, not ours, and compounds into a system no competitor can pull out of your VPC.

02 / One queue for HPC + AI

Any job gets the right hardware — no partition routing.

Classical HPC and AI-for-bio on one queue — structure prediction, docking, MD, RBFE, cryo-EM refinement, genomics pipelines, plus model training and inference. Karpenter picks the instance family and GPU. You never write a partition name.

03 / Your AWS, your IP

Your data never leaves your VPC.

Pre-IND structures, tilt-series, and genomic data are program IP. They stay in your IAM perimeter — not on our infra, not routed through our control plane. You see every dollar on your AWS bill, and your Savings Plans apply.

Three ways in

Console, CLI, or REST — same cluster, your call.

Submit from the web console if you'd rather not touch Slurm. SSH in and sbatch if you live in the terminal. Wire the REST API into your orchestrator if you automate everything.

01 / Web console

Dashboard — submit, monitor, review.

Browse the template catalog, fill the form, hit submit. Live logs, per-job cost stamp, run history. For scientists who'd rather not write sbatch.

Launch the console →

02 / CLI

sbatch fep-campaign --network edges.csv

Slurm-native. SSH in and submit the way you would on any HPC system — same sbatch, squeue, sacct, scancel. For people who already live in the terminal.

CLI quickstart →

03 / REST API

POST /v1/jobs/submit

Wire Clusterra into your in-house pipeline, Airflow DAG, or Slack bot. JSON in, JSON out. For platform teams that want Clusterra as the compute backend.

API reference →

Benchmarks

Full-pipeline cost on the workloads you actually run.

Spot pricing, checkpoint-restart on, us-east-1. Reproducible benchmarks on public datasets — fork the repo, verify the numbers.

Workload
Cost
Reference
CADD hit-discovery funnel — structure → screen → affinity → MD
OpenFold3 + Uni-Dock 1K + Boltz-2 rescore + GROMACS 1ns · L40s spot
$1.1673 min, 1,000 compounds
EGFR ATP site (7TU3) · Enamine Diverse 1K · OpenFold3 TM 0.984 vs crystal
OpenFE RBFE campaign — TYK2, 9-edge network
OpenFE 1.11.1 · 11 λ-windows · 5 ns · A10G spot
~$1009-edge, n=3 · ~$11/edge
TYK2 canonical benchmark, MUE 0.38 kcal/mol vs experiment
Cryo-ET STA — HIV-1 Gag subtomogram averaging
AreTomo2 + RELION-5 tomo refine · 4× T4 spot
~$2.30to 3.99 Å
EMPIAR-10164 · gold-standard FSC 3.99 Å
AlphaDIA proteomics cohort
timsTOF Ultra 2 · HeLa 200ng · Mann lab library
$0.05per sample, 13 min
8,591 proteins, 105,414 precursors · cohort runs in parallel
Sarek + Parabricks 30× WGS
Parabricks GPU alignment + DeepVariant · g5 spot
$2.59per sample, 2h 32m
GIAB NA12878 · SNP F1=0.9966 / INDEL F1=0.9934

See all benchmarks →

Beyond raw AWS · built by ex-AWS HPC

AWS sells you the parts. We hand you the working instrument.

Batch, PCS, and ParallelCluster are those parts pre-sorted but still un-assembled — we know, we helped build them. We deliver the cluster already built, validated, and running — in your own account.

01 / Day-one cluster

Working on day one — not a kit to assemble.

Batch, PCS, and ParallelCluster are kits: someone still has to design the cluster, build the containers, wire up Slurm, and tune the GPUs before a single job runs correctly. That’s a dedicated headcount’s worth of HPC/DevOps work. We hand it to you already built and validated — a managed Slurm cluster running right the day we turn it on.

02 / Survivable spot

Spot GPUs at ~60% off — without the dead runs.

Spot GPU is ~60–70% cheaper than on-demand, but on raw AWS a long-running job dies the instant the instance is reclaimed — so most teams never use it. We catch the 2-minute reclaim warning, drain the job cleanly, and resume it. The cheap compute becomes usable for the long jobs that actually need it.

03 / Costed & reproducible

Every run is costed and reproducible — automatically.

On raw AWS you get one monthly bill and no record of how a result was produced. We stamp every job with its dollar cost as it runs — spend visible per experiment, per user, per project — and capture the exact pinned container digest and script behind each run, so any result can be reproduced or re-run months later without rebuilding the environment.

All of that — the day-one cluster, survivable spot, costed reproducible runs, and your own AWS/NVIDIA credits paying the compute — for less than a tenth of the hire it would take to build it yourself:

Pricing

Pick your plan by how much you run.

Every plan is the full platform. Plans differ only by how much management credit is included.

Solo
$99/mo per cluster

For one scientist getting started. Run real campaigns on a proper GPU cluster without committing to anything bigger.

Included credits$99/mo≈ 100 GPU-hrs
Rate$0.40/GPU-hr + $0.04/vCPU-hrmetered from your credit
OverageSame ratepast your included credits

Get started →
Most popular
Team
$499/mo per cluster

For a team running every week. Pay $499, get $999 of management credit — about 2× the value.

Included credits$999/mo≈ 1,000 GPU-hrs
Rate$0.40/GPU-hr + $0.04/vCPU-hrmetered from your credit
OverageSame ratepast your included credits

Get started →
Org
$1,499/mo

For running hard, with zero math. One flat price covers everything you run — it never changes.

Included creditsUnlimitedone flat fee
RateNoneno metering, ever
Overagenothing to meter

Talk to us →

Against any of those, a $99–$1,499/mo management fee is a rounding error — and not a license, a CRO, or a headcount you'd otherwise carry.

Questions, answered

What exactly am I paying Clusterra for?

For running your cluster — the managed software, the validated scientific stack, spot survival, and provenance. That is the only thing you pay us. The compute itself (GPUs, CPUs, storage) is billed separately and directly by AWS in your own account, at AWS prices — we never touch or mark up that bill.

Which plan should I pick, and how does it bill?

One scientist, light or occasional runs → Solo ($99/mo, includes $99 of management credit). Running most weeks → Team, the best value ($499/mo, includes $999 of credit — ~2×). Running hard and want one predictable bill → Org ($1,499/mo flat, unlimited, no metering).

“Management credit” is prepaid Clusterra fee — not compute, and not your AWS credits. Solo and Team meter it at one uniform rate, $0.40/GPU-hr + $0.04/vCPU-hr; once the credit is used up you keep paying that same rate, nothing changes. Hours vary by instance; AWS bills the compute separately and directly, at AWS prices.

Can I use my AWS & NVIDIA credits?

Yes — this is a real advantage of BYOC. Because compute runs in your own account, your AWS Activate and NVIDIA Inception credits pay for it directly. A SaaS platform that bills you cannot apply those credits — they'd expire unused. Many teams run for months on credits alone; the management fee is the only thing you pay us.

What's the pilot?

A free 30-day pilot: we stand up your cluster in your AWS, wire your workflows, and get a real campaign running with you — no charge.

Can I cancel anytime?

Yes — no lock-in. Cancel anytime and changes take effect on your next cycle. Because everything runs in your own AWS account, your data and clusters stay with you.

Start free — no credit card →

Or book a free pilot — we get a real campaign running, no charge.

$ clusterra submit --workflow your-pipeline --target your-data # Your AWS. Your data. Your credits. 30-day pilot.

Scope a pilot on your science.

30 minutes to scope. Free 30-day pilot with your first campaign running in your AWS account — CADD, cryo-EM, or multi-omics, no charge. Then BYOC plans from $99/mo. Three design-partner slots — closes when filled.

Prefer email? hello@clusterra.cloud · or join the community Slack →