LenovoThinkSystem SR675 V3

A 3U server with one or two AMD EPYC processors and room for eight double-width GPUs, from the RTX PRO 6000 Blackwell to the H200 NVL. It's Lenovo's server for when several models or several teams share one machine.

Ask for price and lead time See what it supports

Rack 3U

What it supports

The SR675 V3 is configured to order on a four- or eight-GPU base: these are its limits according to Lenovo's guide. Lenovo Press product guide

Processor1 or 2 AMD EPYC 9005 or 9004Up to 160 cores and 400 W

GPUUp to 8 double-widthRTX PRO 6000 Blackwell, H200 NVL, L40S or L4

MemoryUp to 6 TB24 DDR5 slots · 6400 MHz with EPYC 9005

StorageE1.S, E3.S or 2.5" NVMeDepends on the GPU base · M.2 boot in RAID

NetworkOCP 3.0 and up to 6 PCIe 5.0ConnectX-7 at 400 Gb/s or BlueField-3, with GPU Direct

Form factorRack 3UBase for 4 or 8 double-width GPUs

Power: Up to four 1,800, 2,400 or 2,600 W power supplies, Titanium or Platinum
Cooling: Air, with five dual-rotor fans
Management: XClarity Controller 2, with the Platinum licence included
Operating systems: RHEL, SLES, Ubuntu, Windows Server and VMware ESXi
Warranty: 1 or 3 years depending on the model, with an on-site engineer next business day

One machine for several models and several teams

With eight RTX PRO 6000 cards you can run the internal assistant's model, a smaller one for classification or data extraction and a third for testing, each on its own GPUs. For models that need more memory per GPU, it takes the 141 GB H200 NVL. The cards communicate over PCIe: if you'll be training large models split across all of them, an HGX such as the SR680a V4 does the job much better.

What we do

We deliver it with RHEL, Ubuntu or SLES and integrate it into Kubernetes or OpenShift with the GPU Operator, so the cluster allocates GPUs to projects and teams without anyone logging into the machine, with vLLM for the models in production. The 400 Gb/s slots with GPU Direct let the GPUs read from shared storage without going through CPU memory, usually Ceph: we explain how in AI inference with Ceph, OpenStack and Kubernetes.

Supported GPUs

GPU	Memory	How many
NVIDIA RTX PRO 6000 Blackwell	96 GB	8, or 4 on the four-GPU base
NVIDIA H200 NVL	141 GB	8, or 4 on the four-GPU base
NVIDIA L40S	48 GB	8, or 4 on the four-GPU base
NVIDIA L4	24 GB	8, or 4 on the four-GPU base

The RTX PRO 6000 and the H200 NVL can only be ordered with AMD EPYC 9005 processors.

What your data centre needs

It takes 3U and has up to four power supplies from 1,800 to 2,600 W; the SR675i V3, with its eight GPUs, uses four 2,600 W units at 230 V. It's air-cooled with five fans, but eight GPUs of up to 600 W push a lot of heat out of the back, so we go through your power per rack and cold aisle with you before the configuration is agreed.

SR675i V3: the same machine in a fixed configuration

Lenovo also sells the SR675i V3, with eight RTX PRO 6000 cards already fitted, ordered as a single package with a three-year warranty. The processors, GPUs, memory, drives and network can't be changed.

Processor: 2 AMD EPYC 953564 cores each
GPU: 8 NVIDIA RTX PRO 6000 Blackwell96 GB each
Memory: 1.5 TB24 × 64 GB DDR5 modules at 6400 MHz
Storage: 2 × 3.84 TB E3.S NVMePlus 2 × 960 GB M.2 for the OS
Network: 4 × 400 Gb/s and 2 × 200 Gb/s portsNVIDIA BlueField-3 B3140H and B3220
Power: 4 × 2,600 W power suppliesTitanium, at 230 V

Compared with the rest of the family

Model	What it has	When to choose it
ThinkSystem SR650a V4	Up to 4 RTX PRO 6000 · Intel Xeon 6	The entry point: one to four GPUs in 2U, with Intel.
ThinkSystem SR680a V4	NVIDIA HGX B300 · 8 GPUs	Eight B300s linked by NVLink, for training and serving large models.
ThinkSystem SR680a V3 with B200	NVIDIA HGX B200 · 8 GPUs	The previous HGX, also air-cooled, when lead time or budget comes first.
ThinkSystem SR780a V3	NVIDIA HGX B200 · liquid-cooled	The same eight B200s in 5U, if your data centre has a water loop.
ThinkSystem SR665 V3	4 RTX PRO 4500 · AMD EPYC	Lenovo's entry point for inference, with the Hybrid AI 241 configuration.

See all Lenovo AI and GPU servers

What the delivery includes

From the factory

Server built to your configuration by Lenovo
1- or 3-year warranty depending on the model, with an on-site engineer next business day

From us

GPU, memory and network sizing based on your model
Rack mounting, XClarity and up-to-date firmware
Operating system, NVIDIA drivers and the AI stack
Testing with your workload before it goes into production

If you need it

24×7 on-call support with SLA
Lenovo warranty upgrades: 4- or 2-hour response, or one or two extra years

How we build it

What runs on top of these servers, explained in detail:

RAG over your own documents with Docling AI inference with Ceph, OpenStack and Kubernetes Rocky Linux for AI, with NVIDIA drivers and CUDA ready All our AI services

Support and training for your team

Independent support from the people who installed it Local LLM and GPU infrastructure course Deep learning course with PyTorch and TensorFlow Red Hat OpenShift training

Questions about the ThinkSystem SR675 V3

How does it differ from the SR650a V4?

The SR650a V4 takes up to four GPUs in 2U with Intel Xeon 6; the SR675 V3 goes up to eight in 3U with AMD EPYC. If four GPUs are enough, the SR650a uses less rack space; if the machine will be shared by several teams or models, the SR675.

Which processor do I need for the RTX PRO 6000?

AMD EPYC 9005. Lenovo only supports the RTX PRO 6000 and the H200 NVL with that generation; with EPYC 9004 you can fit the L40S or the L4.

What warranty does it come with?

Lenovo sells it with one or three years depending on the model, with an on-site engineer next business day, and the SR675i V3 with three. It can be upgraded to 4- or 2-hour response, and on top of that we can provide our 24×7 on-call support with SLA.

Is the SR675 V3 right for what you're building?

Tell us which models you want to run and how many teams will use them, and we'll come back with the configuration, the price and a confirmed lead time.

Ask for price and lead time WhatsApp +34 91 198 02 43