See where your GPU cloud stands on all 143 ClusterMAX criteria.

A free CLI you run on your own nodes. It marks every criterion Meets, Partial, Gap or Unknown, lists the top fixes, and keeps bad GPUs out of the pool. Nothing is sent to us.

Install and run

pipx install git+https://github.com/predgeAI/cluster-readiness predge-ready scan
predge-ready scan --demo broken · 3-node demo clusterReadiness index 18.9% of 143 criteria · Meets 7 · Partial 40 · Gap 3 · Unknown 93gpu-h100-07: XID 79, a GPU fell off the busfailgpu-a40-03: reports an A40, an H100 was expectedfailFailing nodes are drained automaticallyfailGPU and node metrics exporters runningfailReal output of the built-in demo (recorded fixtures), not a real cloud.

What you get

Readiness report

Each of the 143 ClusterMAX criteria marked Meets, Partial, Gap or Unknown, a score per category, and the top fixes first.

GPU health checks

Continuous checks for XID errors, ECC, throttling, temperatures, InfiniBand and NCCL. Bad nodes are drained automatically in Slurm or Kubernetes.

Prometheus metrics

Every check is exported as a metric, so it shows up in the dashboards and alerts you already run.

Signed reports

Sign a report with ed25519 and anyone can check it with predge-ready verify. Runs locally and is open source.

Commands

# install pipx install git+https://github.com/predgeAI/cluster-readiness# readiness report for this cluster predge-ready scan# see a report on a deliberately broken demo cluster predge-ready scan --demo broken# health checks; drain bad nodes in Slurm or Kubernetes predge-ready health --drain# check a signed report predge-ready verify report.json

github.com/predgeAI/cluster-readiness

Pricing

Self-scan

Free

Open source. Run it on your nodes as often as you like.

Readiness review

$2,500 from, one-off

We turn your scan into a fix plan and the ClusterMAX submission package.

Continuous readiness

$1,000 / month per cloud

Health monitoring plus a monthly re-scan. Smaller clouds from $300/month.

Next ClusterMAX windows

Dates are set by SemiAnalysis and may move.

  • ClusterMAX 3.1: around 1 Dec 2026 (tentative)

  • ClusterMAX 4.0: early to mid 2027

Our GPU checks already caught a swap

On a GPU rental settled on Arc mainnet, our probe saw that the host had swapped the A100 for an A40. That day was refunded to the buyer.

See the transactions

What we do not promise

  • No rating tier. ClusterMAX ratings are decided by SemiAnalysis alone; the scan shows where you stand and what to fix.

  • Unknown means the scan could not check it from the node. Those items need evidence from your team.

Independent tool, not affiliated with SemiAnalysis. ClusterMAX is a trademark of SemiAnalysis.

Run the scan first.It is free and stays on your nodes. Book a review if you want help with the gaps.