Lenovo · Servers and AI infrastructure

Lenovo ThinkSystem and ThinkAgile Servers

GRECO specifies, procures, integrates and manages Lenovo ThinkSystem and ThinkAgile infrastructure. The hardware is chosen for the workload and for the power and cooling envelope of the site — which, on a GRECO campus, is engineered for liquid cooling and heat recovery from the start rather than retrofitted toward it.

Dell PowerEdge is covered separately — GRECO is not tied to one vendor, and the choice is made per project.

The families we deploy

Specifications live in Lenovo’s own product guides, which are revised in place and stay current. Each family below links to its guide.

ThinkSystem SR — rack servers

General compute and virtualisation

The volume of a campus build: two-socket rack servers for virtualisation, storage nodes and the control plane that sits alongside the accelerated tier.

Includes: SR630 V4 · SR650 V4 · SR635 V3 · SR655 V3 · SR850 V4 · SR860 V4

Lenovo product guide
ThinkSystem SD — dense and GPU systems

AI training, inference and HPC

High-density nodes, including direct water-cooled designs. These are the systems that make a liquid-cooled hall worth building, and the tier GRECO sizes the power and heat-recovery loop around.

Includes: SD650 V4 · SD650-N V3 · SD665 V3 · SD665-N V3 · SR675 V3 · SR680a V4 · SR650a V4

Lenovo product guide
ThinkSystem SE — edge servers

Low-latency compute outside the core

Short-depth, ruggedised systems for edge sites where inference has to happen close to the load rather than in a distant region.

Includes: SE350 V2 · SE360 V2 · SE450 · SE455 V3

Lenovo product guide
ThinkAgile — integrated and hyperconverged

Pre-validated infrastructure

Factory-integrated stacks where a validated platform matters more than component-level choice — the same argument as a rack-scale AI system, applied to general infrastructure.

Includes: ThinkAgile HX · ThinkAgile VX · ThinkAgile MX

Lenovo product guide
ThinkSystem ST — tower servers

Site and branch infrastructure

Tower systems for site offices, control rooms and branch locations that need local compute without a rack.

Includes: ST50 V3 · ST250 V3 · ST650 V3

Lenovo product guide
Storage and networking

The rest of the stack

Storage arrays and switching specified alongside compute, so the fabric and the capacity tier are sized against the same workload rather than bolted on afterwards.

Includes: ThinkSystem DG / DM / DE storage · ThinkSystem switching

Lenovo product guide

Model configuration guide

Lenovo ThinkSystem SR675 V3

Compare GPU configurations, compatible parts, cooling requirements and complete-system pricing factors.

Read the model guide (English)

Lenovo ThinkSystem pricing and configuration planning

A ThinkSystem family name is a starting point for selection, not a complete purchase specification. Use the linked Lenovo product guide to confirm the exact model, generation and supported options, then compare a matching bill of materials.

  • Start with the workload: AI training or inference, virtualisation, storage or an edge deployment. Define processor, system memory and accelerator requirements, and verify GPU compatibility in the guide for the selected model.
  • Include storage and networking: drive type and capacity, boot configuration, network adapters and the fabric connecting the systems. Check software and licence costs separately, including the selected ThinkAgile stack where relevant.
  • Check the site: rack dimensions, power budget and the cooling method required by the chosen configuration. Rack integration, cooling equipment, installation and ongoing hosting should have their own cost lines.
  • Match commercial terms: quantity, warranty and support duration, currency, shipping, taxes and support coverage in the United States or Canada. Availability and delivery timing require confirmation for the specified system.

Comparing vendors? Use the same workload, bill of materials and deployment scope when reviewing Dell PowerEdge GPU server pricing considerations.

Lenovo infrastructure: common questions

The dense ThinkSystem SD line, including the direct water-cooled designs, is what a liquid-cooled hall is built around — that tier sets the rack power and the temperature of the heat you can recover. Two-socket SR rack servers carry the virtualisation, storage and control plane alongside it, and SE edge systems handle inference that has to sit close to the load.

Yes. GRECO specifies, procures, integrates and manages infrastructure from both, and the choice is made per project. Hardware is selected for the workload and for the power and cooling envelope of the site rather than for a single vendor relationship.

Because a server you cannot power is worth nothing. Across much of North America a new large load waits five to seven years for a utility interconnection. GRECO campuses generate behind the meter, so the constraint moves from a queue position to a construction schedule — and the vendor decision becomes a specification question rather than a scheduling one.

Yes, and the water-cooled designs make it easier. Direct liquid cooling returns heat at a higher and steadier temperature than air, which is what allows it to be metered into a district energy network at roughly 50 °C instead of being rejected to the atmosphere.

The exact model and generation, processors, supported accelerators, memory, storage, networking and support term determine the system configuration. Compare matching bills of materials and separate hardware from software licences, rack integration, cooling and hosting. For deployments in the United States or Canada, confirm currency, delivery costs and support coverage at the destination before comparing totals.

Talk to us about a configuration

Configurations are scoped per project, against the workload and the power available on the site. The useful first conversation is what you need to run and when it has to be live.

→ Start a conversation→ AI compute capacity→ How the power is generated