L10 vs L11 integration

L10 vs L11 side by side

Detail on each level: L10 server integration and L11 rack integration.

The full L1–L12 scale

What L11 testing catches that L10 can't

Every L10 server can pass its own tests and the rack can still fail. L11 testing finds the faults that only exist once servers are connected:

• Cabling errors: a link in the wrong port, or the wrong cable type, only shows up once every node is connected.
• Marginal links: optics and cables that pass a link check but throw errors under collective load.
• Fabric behaviour: collective performance depends on topology and switch configuration, which a single server can't test. See why a GPU fabric runs at half bandwidth.
• Power: PDUs and feeds under the whole rack's peak draw.
• Thermal: hot spots, and airflow or coolant problems, with every node at load.
• Configuration drift: a node with different firmware or BIOS settings from its neighbours.

Found in the factory, these are fixed before freight, installation and site time are spent.

Choosing between L10 and L11

Choose L10 when:

• Racks, power, cooling and the network fabric already exist on site.
• Your team racks, cables and validates hardware itself.
• You're adding servers to an existing cluster.
• You're deploying a handful of servers, not whole racks.

Choose L11 when:

• You're standing up new racks or a new site.
• The racks are liquid-cooled or above roughly 30–50 kW.
• There's a fixed go-live date and little room for on-site debugging.
• You don't have an in-house integration team, or it's already committed elsewhere.

Talk about L12 when a deployment spans several racks on one fabric and needs cluster-level acceptance testing.

Cost and timeline: what changes

L11 adds cost for rack hardware, switching, optics and cabling, PDUs, cooling distribution, rack-level test time, crating and freight. It removes cost too: on-site integration labour, the time between delivery and production, and the risk of debugging a rack in your own data center. Which way it nets out depends on whether you already have the team and infrastructure to do the integration yourself.

Either way, the timeline is usually set by the longest-lead component rather than by assembly. One reported build waited 14 weeks for a single part. Ask for itemised quotes, so the two levels can be compared line by line.

OEM, ODM and integrator: who works at which level

• ODMs design and manufacture platforms, and commonly ship barebones at L6 to brands and integrators. Hyperscalers buy direct from ODMs.
• OEMs sell branded servers, often configured to L10, and many run their own rack integration programmes.
• System integrators take ODM and OEM platforms and build them to L10 or L11 for a specific customer.

TORmem works as a build-to-order integrator: it sources platforms from Tier-1 ODMs and OEMs and integrates them at L10 or L11 to each client's specification. See ODM and OEM sourcing.

Frequently asked questions

What's the difference between L10 and L11 integration?

L10 delivers individual servers that are fully configured and tested. L11 delivers a complete rack: those servers plus switches, cabling, power distribution and cooling, assembled and tested together before shipping.

When should I buy L10 instead of L11?

Choose L10 when you already have racks, power, cooling and a network fabric, when your team integrates hardware itself, or when you're adding servers to an existing cluster. Choose L11 for new racks, liquid-cooled designs, or fixed go-live dates without an in-house integration team.

Is L11 integration more expensive than L10?

The rack's bill is higher, because L11 adds switching, cabling, power distribution, cooling, rack-level testing and freight. It can still cost less overall by removing on-site integration labour and shortening the time from delivery to production. Compare itemised quotes rather than headline prices.

What does L11 testing catch that L10 testing can't?

Faults that only appear once servers are connected: miscabled or marginal links, fabric and collective-communication performance, power behaviour at full-rack peak, thermal hot spots under load, and configuration drift between nodes.

Can we do L11 integration in-house?

Yes, if you have the space, tooling, test suites and staff. You'll need the rack design, cabling plan, power and cooling validation and fabric testing that an integrator would otherwise provide, and you carry the schedule risk of problems found on site.

Where does responsibility sit if a component fails at each level?

With L10, the supplier is typically responsible for each server as delivered, and your team for how the servers are racked, cabled and configured. With L11, the integrator is typically responsible for the rack as a tested whole up to handover. Agree the boundary, warranty and RMA process in the contract.

What is L12, and when do you need it?

L12 connects multiple L11 racks into a cluster, adding cross-rack cabling, fabric configuration and cluster-level testing, and sometimes management software. It's relevant when a deployment spans several racks on one fabric.

How do L-levels map to OEMs, ODMs and integrators?

ODMs commonly ship barebones at L6 to brands and integrators. OEMs sell branded servers, often at L10, and some run rack programmes. System integrators build to L10 or L11 for a specific customer, using ODM and OEM platforms.

Do vendors define L10 and L11 the same way?

Not always. The scale is an industry convention rather than a standard: some vendors use L10 for a mechanical rack shell, and NVIDIA's supply chain uses L10 for finished compute trays. Ask each vendor for an itemised scope.