All from Lenovo

Lenovo ThinkSystem SR680a V4

Lenovo ThinkSystem SR680a V4 - AI Server
This item cannot be returned by withdrawing from a distance sales contract once it has been unpacked. For more information, see the Terms and Conditions.
Pre-sale price 65,990,000,-
without VAT  54,537,190,-
Add to the list

Lenovo ThinkSystem SR680a V4

ℹ️

This product is intended solely for businesses, public sector bodies, and educational institutions. Completing an order on the e-shop does not constitute a binding contract, but serves as an inquiry only.

The sale of this product is subject to export, sanction, and other legal restrictions. Before we can accept your order, we must verify the buyer, end-user, delivery country, and intended use. We can only accept the inquiry after successful authorisation from the supplier.

If the required conditions are not met, we reserve the right to reject the inquiry or cancel the order without any entitlement to delivery of the goods.

If you are interested in purchasing AI solutions or accelerators, you can also write to us at [email protected]. We will then prepare a custom quote and individual contract. For more details, please see our terms and conditions (clauses 3.33 and 3.34).

Lenovo ThinkSystem SR680a V4: AI factory for trillion-parameter models

The largest language models no longer have to run exclusively on hyperscaler infrastructure. Lenovo ThinkSystem SR680a V4 brings the power of a fully fledged AI factory into a single 8U server under your complete control. Eight NVIDIA B300 GPUs connected via NVLink and 2.3TB of HBM3e memory let you train, fine-tune, and run models ranging from hundreds of billions to a trillion parameters, while serving thousands of users simultaneously. Your data remains secure within your company, latency is minimal, and costs are predictable. Whether you are building an enterprise AI platform, a research hub, or generative services for your customers, the SR680a V4 is the ideal choice. This configuration was created by Alza as part of the Lenovo AI Express programme.

  • 8× NVIDIA B300 GPU (HGX B300 NVL8) — a total of 2304GB of HBM3e memory (288GB per GPU) connected via NVLink for training, fine-tuning, and inference of the largest language models.
  • Lenovo ThinkSystem SR680a V4 platform — air-cooled 8U GPU server featuring 8 OSFP 800Gb/s ports for seamless compute cluster interconnectivity.
  • 2× Intel Xeon 6776P — 2× 64 cores / 128 threads, 2.3GHz clock speed (up to 3.9GHz turbo boost), and 336MB cache per processor.
  • 2048GB DDR5 ECC Registered (RDIMM) — 32 modules of 64GB each across 32 slots; 6400MT/s modules operating at 5200MT/s when fully populated.
  • 8× 3.84TB NVMe SSD U.2 — enterprise-grade drives with PCIe 5.0 interface (30.72TB in total), complemented by 2× 960GB M.2 NVMe for the operating system.
  • Delivery within 25 days — configuration from the Lenovo AI Express programme; RAM and SSD capacity can be customised upon request.
8x
NVIDIA B300 GPU
2304GB
total GPU memory (HBM3e)
2048GB
DDR5 ECC Registered
128cores
2× Intel Xeon 6776P

SYSTEM BREAKDOWN

Component details

8× NVIDIA HGX B300 NVL8
Two RTX PRO 6000 Blackwell Server Edition cards feature 96GB of memory per card, providing 192GB in total. This is ample for models with tens of billions of parameters, long contexts and concurrent requests. The passive cards are cooled by high-performance server fans, and each features a PCIe 5.0 x16 connection and up to 600W power delivery. They are ideal for inference, model fine-tuning, generative AI and visualisation. MIG technology lets you partition each GPU into up to 7 isolated 34GB instances, allowing the server to serve multiple teams or services at once.
2× Intel Xeon 6776P
Two Granite Rapids generation processors deliver a combined 128 cores and 256 threads, with a 2.3GHz base clock and up to 3.9GHz turbo. Each offers 8 memory channels, a PCIe 5.0 interface with 88 lanes and built-in AMX, QAT, DSA, IAA and DLB accelerators. The processors prepare data, manage the system and feed the GPUs without delay, ensuring the performance of the eight B300s is never left idle.
1536GB DDR5 ECC Registered + 2× 3.84TB NVMe
The configuration includes 24 modules of 64GB TruDDR5 RDIMM 6400MHz, totalling 1536GB RAM. Use them for data preprocessing, vector databases and RAG retrieval layers. For models and datasets, two 3.84TB E3.S SSDs with an NVMe PCIe 5.0 x4 interface (7.68TB total) are connected without throughput limits. The operating system runs on two 960GB M.2 NVMe SSDs.
ThinkSystem SR675 V3
The SR675 V3 is a 3U server designed specifically for GPUs. It is supported by five dual-rotor fans, redundant power supplies (four 2600W 80 PLUS Titanium hot-swap PSUs) and the XClarity Controller 2 for remote management, including KVM and virtual media. Security is enhanced by TPM 2.0, Secure Boot and a chassis intrusion sensor.

CPU

2× Intel Xeon 6776P

Modern processors from Intel

Two Granite Rapids generation processors deliver a combined 128 cores and 256 threads, with a 2.3GHz base clock and up to 3.9GHz turbo. Each offers 8 memory channels, a PCIe 5.0 interface with 88 lanes and built-in AMX, QAT, DSA, IAA and DLB accelerators. The processors prepare data, manage the system and feed the GPUs without delay, ensuring the performance of the eight B300s is never left idle.
The Xeon 6776P is part of the Lenovo AI Express programme, making its availability better than standard configurations.

PLATFORM

Powered by LENOVO

ThinkSystem SR680a V4

The SR680a V4 is designed for 24/7 continuous operation under full load. Six front fans cool the processors, memory and storage, while another 15 rear fans extract heat from the GPU section, keeping the airflow moving from front to back. Power is supplied by eight 3800W Titanium PSUs. Management is handled by the XClarity Controller 3 with a Premier licence, including remote KVM, virtual media, and FIPS 140-3 and CNSA modes. Security is rounded off by TPM 2.0 and Secure Boot.
The server weighs up to 124.7kg. We recommend using a lift for server maintenance.

Key technologies

What makes the Lenovo ThinkSystem SR680a V4 exceptional?

8× NVIDIA B300 SXM6

Each GPU delivers memory bandwidth of up to 7.7TB/s.

2304GB GPU memory

Combined capacity of all cards for large models.

128 Xeon cores

Granite Rapids delivers data to the GPUs without latency.

2048GB DDR5 and 32 slots

Increase capacity up to 4TB.

8× 3.84TB NVMe SSD U.2

Enterprise SSD with PCIe 5.0 interface.

Eight 3800W power supplies

Power supply for non-stop operation.

PCIe 5.0

Full throughput for both GPUs and network adapters.

8× 800Gb/s

OSFP ports for cluster interconnect.

Connectivity

Ports and slots

Front panel

  • 8× OSFP 800Gb/s
  • 4× USB 3.0
  • 1× VGA
  • 1× RJ-45 1GbE for XCC3 management
  • Slots: 4× PCIe 5.0 x16 FHHL for expansion cards and 1× OCP 3.0 PCIe 5.0 x16
  • The configuration includes NVIDIA BlueField-3 B3240 (2× 400Gb/s) and Intel X710-T2L (2× 10GBase-T) network cards
Lenovo ThinkSystem SR680a V4 server

Dimensions

8U rack format

The server occupies 8U in a standard 19" rack. Operating ambient temperature ranges from 10–35°C. Before installation, please verify the rack depth, as well as the power and cooling capacity in the data centre.

351mm
height
447mm
width
924mm
depth

Specifications

Technical overview

Type8U rack server
Platform    Lenovo ThinkSystem SR680a V4 (7DMK)
CPU2× Intel Xeon 6776P, 64 cores / 128 threads, 2.3GHz (3.9GHz turbo), 336MB cache, 350W
GPU8× NVIDIA B300 SXM6 (HGX B300 NVL8), 288GB HBM3e per GPU, 2304GB total
RAM2048GB (32× 64GB TruDDR5 RDIMM), 32 slots, up to 4TB, 5200MT/s at 2 DIMMs per channel
Storage8×3.84TB NVMe U.2 PCIe 5.0, 2×960GB NVMe M.2
Network8×OSFP 800Gb/s, NVIDIA BlueField-3 B3240 2×400Gb/s, Intel X710-T2L 2×10GBase-T
Power supply8×3800W 80 PLUS Titanium
Management and securityTPM 2.0, Secure Boot
Form factor447×351×924mm (W×H×D)

Warranty and service

Premium service guarantee

Lenovo ThinkSystem SR680a V4 server comes with a 3-year warranty featuring 24×7 service and a guaranteed 4-hour response time. Submit requests at any time, including nights, weekends, and holidays, and a Lenovo technician will respond within 4 hours. You also benefit from the 36-month Keep Your Drive service, meaning faulty drives stay with you, keeping sensitive data inside your company. This offers significantly more protection than standard 9×5 service for business deployments where every hour of downtime costs money.

4hours
response to service request
36months
warranty
24×7
technical support

Request a quote for Lenovo ThinkSystem SR680a V4

Let us know your deployment purpose, required configuration, and deadline – we will get back to you with an offer.

Nebo nám napište přímo na [email protected]

Děkujeme! Vaši poptávku jsme odeslali, ozveme se co nejdříve.

Questions

Frequently asked questions

Who is Lenovo ThinkSystem SR680a V4 designed for?
It is built for large companies, research centres, and AI service providers training or running the most demanding models. It is ideal for generative AI, fine-tuning, scientific simulations, and rendering.
What size of large language model can I run on this server?
According to the Lenovo AI Express classification (Size L), it is designed for models ranging from approximately 400 billion to a trillion parameters, supporting thousands of users at speeds of 30 or more tokens per second. Actual performance depends on quantization, context length, and the server application used (e.g. vLLM, TensorRT-LLM).
How much GPU memory does the system offer?
2304GB HBM3e (8×288GB). An NVLink interconnects the eight GPUs, allowing models to scale across them highly efficiently.
Do I need a special power supply?
Yes. The server has 8×3800W power supplies (80 PLUS Titanium) and requires high-voltage power (200V and above). Verify your rack capacity and room cooling before installation.
What is included in the delivery and how quickly will it arrive?
The server comes with power supplies, cables, rails, and handles. The Lenovo AI Express program enables delivery in as little as 25 days from the distributor order (subject to meeting all requirements, such as the EUC document). The sales department will confirm the exact date.
Can the server be upgraded?
Yes. The platform supports up to 4TB RAM, higher-capacity drives, and additional network cards in four PCIe slots. The number of GPUs is fixed (8×B300). Consult us regarding upgrades to ensure compatibility and maintain your warranty.

Calculators

What can you run on this server?

Interactive calculations based on the 2.3TB GPU memory of this server: see what size of large language models fit and how fast they respond.

⚡Živá kalkulačka: Jaký model rozběhneteInteraktivní výpočet se načte na webu (engine z CDN).
⚡Živá kalkulačka: Jak rychle mluvíInteraktivní výpočet se načte na webu (engine z CDN).
Lenovo ThinkSystem SR680a V4 server
Lenovo ThinkSystem SR680a V4 server
Lenovo ThinkSystem SR680a V4 server

Specifications

Processor

Processor series Intel Xeon
Processor model number Intel Xeon 6357P
Processor frequency 2.3 GHz (2,300 MHz)
Number of processor cores 64 ×
Processor cache 336 MB
Core boost frequency 3.9 GHz (3,900 MHz)
Max TDP 350 W
CPU features HyperThreading, Virtualization Support, Multi Threading, No graphics chip, Turbo Boost

Motherboard

Socket LGA 4710
Slots for additional cards PCI Express x16, M.2
Design RDIMM

disk

Storage Type SSD
Storage capacity (total) 32.6 TB
SSD Capacity 32.6 TB
Internal interface U.2 (PCIe 5.0 4x NVMe)
Max. number of disks 10 pc(s)
Disk size 2.5"
Hot Plug Hot Plug

Colour and design

Colour Black
Wardrobe size Rack
Front panel location front

PSU

PSU 30,400 W

Memory

Size of operational RAM 2,048 GB
Memory frequency 6,400 MHz
Memory type DDR5
Maximum RAM capacity 4,096 GB
Number of RAM slots 32 pc(s)
Number of installed slots 32 pc(s)

Outputs

USB 3.2 Gen 1 (USB 3.0) 4 ×
Graphics VGA D-SUB
Other NVIDIA ConnectX-8 SuperNIC, RJ-45 management LAN port
Optical drive Without Optical Drive

Case

Width 447 mm
Height 351 mm
Depth 924 mm
Weight 124.7 kg

Operating system

Operating system Without Operating System

Product series

PC series Lenovo ThinkSystem

NCNR (Non-Cancellable, Non-Returnable)

Business purchasing method On request with individual contract
NCNR - Typ smlouvy Full
Watch Price/Availability
Best sellers, best servers Servers reviews Cheap servers Cheap servers Lenovo Servers Servers Lenovo RACK - Racks, Servers, Accessories RACK - Racks, Servers, Accessories Lenovo All categories
Previous Other
Code:  TI896a_5AI3
Warranty: 36 months
Links: Producer's Website:
Manufacturer information

Lenovo Technology B.V. organizační složka, Classic 7, Jankovcova 1603/47a, 17000 Praha 7, [email protected], https://www.lenovo.com/cz/cs/contact/; Lenovo (Slovakia) s.r.o., Landererova 12, 811 09 Bratislava, [email protected]


Lenovo Technology B.V. Classic 7, Jankovcova 1603/47a, 17000 Praha 7, [email protected], https://www.lenovo.com/cz/cs/contact/; Lenovo (Slovakia) s.r.o., Landererova 12, 811 09 Bratislava, [email protected], https://www.lenovo.com/

Lenovo is the world's largest computer maker, founded in 1984. Its original name was Legend. Later it was renamed to Legend Novo ("New Legend"), abbreviated Lenovo. In 2005 it took over IBM's computer division, which was among the largest computer makers in the world. It is similarly ambitious in the field of mobile phones, where it would like to surpass Apple and Samsung. As far as this goal goes, they acquired Motorola’s mobile division in January 2014 and have been ramping up production since. 

The company also produces tablets, monitors, televisions, and storage solutions. Overall, they are known for delivering quality products at reasonable prices.

Show more

Related categories