COGNITIV / Servers

The right server for your AI

COGNITIV is installed on Lenovo ThinkSystem and ThinkEdge servers, in your server room or your data centre. We choose the model based on the AI you will use, how many people will use it at once and where it will be installed.

See the catalogue Size my server

Server catalogue

Ten reference models. Filter by use or click a server in the rack.

10 servers

Model	Use	Form factor	Processor	GPU	Memory	Storage
ThinkSystem SR650i V4Reference configuration for COGNITIV	AI inference	2U	2× Intel Xeon 6530P (32 cores)	2× NVIDIA RTX PRO 6000 Blackwell	512 GB TruDDR5 6400	6× NVMe + 2× SATA
ThinkSystem SR675 V3Many GPUs for company-wide inference	AI inference	3U	1–2× AMD EPYC 4th/5th gen.	Up to 8 double-width or 4 SXM5	24× RDIMM TruDDR5	8× SAS/SATA/NVMe 2.5″
ThinkEdge SE350 V2Branches, plant floor or shop	Outside the data centre	1U, half width	1× Intel Xeon D-2700	No dedicated GPU	256 GB	4× 2.5″ hot-swap
ThinkSystem SR685a V3Large generative models	Generative AI at scale	8U	2× AMD EPYC 4th/5th gen.	8× AMD MI300X or NVIDIA H100/H200	24× RDIMM TruDDR5	16× NVMe 2.5″
ThinkSystem SR780a V3Neptune liquid cooling; training too	Generative AI at scale	5U	2× Intel Xeon 5th gen.	8× NVIDIA SXM H100/H200/B200	32× TruDDR5 5600	16× NVMe U.3
ThinkSystem SR650 V3Versatile server for mixed workloads	General purpose	2U	2× Intel Xeon 5th gen. (up to 60 cores)	Up to 8 single-width or 3 double-width	32× TruDDR5	40× 2.5″ or 36× NVMe
ThinkSystem SR655 V3Single processor, good price-performance	General purpose	2U	1× AMD EPYC 4th/5th gen.	Up to 8 single-width or 3 double-width	TruDDR5	40× 3,5″/2,5″
ThinkSystem SR630 V3Maximum density in little space	General purpose	1U	2× Intel Xeon 5th gen. (up to 60 cores)	3 single-width 75 W GPUs	32× TruDDR5	12× 2.5″ or 16× EDSFF
ThinkSystem SR650 V2Previous generation, critical workloads	General purpose	2U	2× Intel Xeon Scalable 3rd gen.	GPU-capable	TruDDR4	Flexible storage
ThinkSystem SR630 V2Previous generation, 1U density	General purpose	1U	2× Intel Xeon Scalable	GPU-capable	TruDDR4	HPC, cloud or HCI

Maximum reference specifications from the Lenovo ThinkSystem and ThinkEdge range. The final configuration is sized to your use case. Lenovo, ThinkSystem, ThinkEdge, Neptune and XClarity are trademarks of Lenovo; product images belong to their respective owners.

How we choose your server

Three questions decide the size. We answer them with you during the assessment.

Which AI model will you use?

A compact model fits on one or two GPUs. Large generative models need eight-GPU servers with high-speed links between them.

How many people at once?

One department or the whole company. The number of simultaneous queries determines how many GPUs and how much memory are needed.

Where will it be installed?

Data centre, technical room, branch office or plant floor. Outside the data centre, a compact, quiet edge unit is the best fit.

From assessment to managed service

We support you in four steps, from the first use case to day-to-day operation.

COGNITIV assessment
Before we start
We assess data, processes, security and team to choose the highest-impact use cases and size the server.
COGNITIV pilot
About 3 months
A real use case running on your server, with your data, ready to go into production.
Rollout and scaling
In phases
We extend it to more departments and agents, adapted to your workflows, with a single point of contact.
Managed service
In operation
Monitoring, updates, model tuning and support so the platform stays secure and up to date.

Lenovo technology included

NVIDIA GPUs

NVIDIA RTX PRO, HGX H200 and B200 GPUs, with NVLink on the high-density models.

Neptune cooling

Closed-loop liquid cooling on the eight-GPU models: lower power consumption and less noise, with no building work in your server room.

XClarity management

Monitoring, updating and provisioning of the server throughout its life cycle, under your control.

Let's size your AI server

Tell us about your case and we will propose the model and configuration.

Talk to a specialist