TerawattIQ
Start your AI factory bring-up Customer login

CLUSTER PASSPORT

Prove production readiness.

Defined tests. Measured thresholds. Recorded exceptions. Versioned acceptance evidence.

CLUSTER PASSPORT

Prove it performs.

Every deployment is validated against defined acceptance criteria and recorded in a versioned Cluster Passport.

Download sample PDF
TERAWATTIQ / CLUSTER PASSPORT

TWQ-B300-1024-REF

✓ CERTIFIED
ACCELERATORS1,024 × NVIDIA B300
VERIFIED2026-09-03 · 22:14 UTC
Fabric / NCCL all-reduce450 GB/s observed455 reference · ≥440 acceptance · PASS
Storage / checkpoint write18.4 TB/s observed18.8 reference · ≥17.5 acceptance · PASS
72-hour fabric soak0 drops observed0-drop acceptance · PASS
Cooling / full synthetic load7.4°C ΔT observed6–9°C acceptance · PASS
GPU burn-in / soak / training / checkpoint PASSRecipe B300-RDMA-1.0 · Firmware / driver state recordedInspect the acceptance record →
SHA-25690d18f8c…e42a7b11Signature valid
DiscoverConfigureTestMeasureRemediateRetestSign

Identity + configuration

ArchitectureTopologyHardware identityConfiguration recipeSoftware versionsFirmware + drivers

Acceptance evidence

Test definitionsThresholdsObserved valuesPass / failExceptionsRemediationRetests

Lifecycle + integrity

Version historyMaintenance eventsRevalidationHashSigner identity + signatureApproval history

A checksum detects a change to evidence; authenticity also requires a trusted signer and verification policy. The interactive demo issues sample evidence, not third-party certification.

Cluster acceptance

A benchmark is a number.
A certification is a promise.

We test the complete system the way customers will use it, then preserve the evidence in a versioned cluster passport.

01

Physical truth

Inventory, topology, cabling, optics, firmware, BIOS, BMC, PCIe, NUMA, and power.

02

Component health

Vendor diagnostics, sustained burn-in, memory, CPU, NVMe, GPU, NIC, and thermal behavior.

03

Fabric performance

Link and error health, RDMA, congestion, NVLink/NVSwitch, collectives, and scaling efficiency.

04

Application readiness

HPL, storage, distributed training, checkpoint/restart, fault injection, and sustained soak.

TERAWATTIQ CERTIFIED

Cluster Passport

Architecture
GPU + fabric + storage profile
Configuration
Versioned deployment recipe
Evidence
Raw output + normalized results
Thresholds
Expected vs. observed performance
Lifecycle
Failures, repairs, upgrades, re-tests
Machine-readableSignedAuditable
DCGM / AGFHCGPU BURNNVLINK / NVSWITCHPCIE / NUMARDMANCCL / RCCLHPL / HPL-AIIOR / MDTESTDISTRIBUTED TRAININGCHECKPOINT / RESTARTSOAK

Build with confidence

Your GPUs should be generating results—not tickets.

Bring your infrastructure. TerawattIQ turns it into certified production capacity and keeps it operating.

See the platform in action