Skip to main content

[ Cluster audit ]

Find out how your cluster holds upbefore your customers do.

A reliability and security audit mapped to the criteria ClusterMAX publishes. Read-only access, a ranked fix list, and a re-test.

Book your auditSee what we test
audit · 4 nodes · 32 gpusexample findings
PASSdriver and firmware versions match on 32/32 GPUs
PASS8h burn-in, 0 XID events
WARNNCCL all-reduce 312 GB/s, design target 360+
FAILfailed node detected in 6m 40s, target 2m
FAILmetrics API readable across tenants
WARN2 security bulletins unpatched > 30 days
fix list → 2 critical · 2 high · re-test booked

[ What we test ]

The phases a 15‑minute auditcan't cover.

We pick up where a config audit stops: under load and under failure.

Audit scopeUnder load and under failure

  • Reliability

    Burn-in, failure injection, and time to drain a failed node.

  • Security

    Tenant isolation, patch currency, and management exposure.

  • Performance

    NCCL, GPU, and storage measured against the design.

  • Read-only access
  • Outbound HTTPS only
  • No inbound ports, no SSH, no root

[ How it works ]

Four steps,one report.

EngagementFindings mapped to ClusterMAX categories

  • 01 · Scope

    A 30-minute call on size, scheduler, and test windows.

  • 02 · Connect

    A read-only collector over outbound HTTPS.

  • 03 · Test

    Performance and reliability runs in your window.

  • 04 · Report

    A fix list ranked by impact, then a re-test.

[ Book ]

Book your audit.

Tell us the cluster size and scheduler, and attach your ClusterMAX CLI output if you have it.

Loading calendar...

ClusterMAX is a trademark of SemiAnalysis. Factryze is not affiliated with or endorsed by SemiAnalysis, and a Factryze audit is not a ClusterMAX rating.