Services / 01

From powered racks to accepted capacity.

Commissioning is where cluster schedules are won or lost. We take the cluster from first power-on through fabric bring-up, burn-in and acceptance, and hand over capacity that runs production workloads.

01 / Where we start

We pick up where facility commissioning finishes. Once power and cooling have passed integrated systems testing (the final facility test that runs power, cooling and controls together under load), we commission the cluster itself. We agree the handoff with your facility commissioning agent so nothing falls between the two.

02 / What we lead

  1. 01

    Commissioning plan

    Test scripts and pass criteria agreed against your contract's acceptance terms before hardware lands.

  2. 02

    Power-on and node bring-up

    Firmware baselines, management network access, operating system and driver images, node by node.

  3. 03

    Fabric bring-up

    Link training, cable and optic fault clearing, topology checks against the rail map (the wiring plan that connects each GPU to its network plane), and fabric-wide performance tests such as all-reduce, the collective operation that training jobs depend on.

  4. 04

    Burn-in

    Sustained compute, memory and network stress at full scale, with node-by-node fault isolation so failures are found now and not in production.

  5. 05

    Fault and parts management

    Every failed part tracked through the vendor replacement process, reinstalled and retested.

  6. 06

    Acceptance

    Acceptance testing against contract terms, and closing the gaps with each vendor until the criteria are met.

  7. 07

    Handover

    Runbooks, as-built records, fault history and an open items list, handed to your operations team.

03 / What you get

  • A commissioning plan agreed before hardware arrives
  • Daily burn-in status and a live fault register
  • Completed acceptance documentation
  • A handover pack your operations team can run from

04 / Where it sits in the delivery

CAPACITY SIGNED 01 Readiness and design validationM1 · DESIGN FREEZE Validated design 02 Build leadershipM2 · INSTALLED AND CABLED Delivery program and reporting 03 Burn-in M3 · BURN-IN COMPLETE Stable cluster through burn-in 04 Acceptance M4 · CUSTOMER ACCEPTANCE Completed acceptance 05 Workload-ready handoverM5 · PRODUCTION LIVE Handover pack and runbooks CAPACITY SIGNED 01 Readiness and design validation Validated design M1 · DESIGN FREEZE 02 Build leadership Delivery program and reporting M2 · INSTALLED AND CABLED 03 Burn-in Stable cluster through burn-in M3 · BURN-IN COMPLETE 04 Acceptance Completed acceptance M4 · CUSTOMER ACCEPTANCE 05 Workload-ready handover Handover pack and runbooks M5 · PRODUCTION LIVE

Phases 3 Burn-in, 4 Acceptance, 5 Workload-ready handover. Closes at M3 Burn-in Complete, M4 Customer Acceptance and M5 Production Live.

05 / When to bring us in

  • Hardware lands in the next 90 days and no one owns commissioning end to end.
  • Burn-in has run for weeks and the fault list is not shrinking.
  • Acceptance has stalled between you and a vendor, and a financing milestone depends on it.
NEXT STEP

Tell us when your hardware lands. We will come prepared.

Arrange an introduction