Affordable Lenovo ThinkSystem GPU Servers: Refurbished AI & HPC Servers with NVIDIA GPUs
- server-parts.eu server-parts.eu

- Feb 25
- 3 min read
Updated: Nov 1
Many Lenovo users are choosing refurbished Lenovo ThinkSystem servers to expand GPU capacity without large investments.
Lenovo ThinkSystem Serves: Save Up To 80%
✔️ 5-Year Warranty – No Risk: Pay Only After Testing
These servers, fitted with NVIDIA GPUs, deliver strong performance for AI and HPC workloads at a fraction of the cost of new systems.
Why Refurbished Lenovo ThinkSystem GPU Servers
Lenovo’s ThinkSystem line offers enterprise reliability, high compute density, and flexible GPU configurations. Refurbished units retain full support for NVIDIA GPUs, making them suitable for AI training, inference, and simulation tasks. Combining refurbished GPUs with refurbished servers allows companies to build capable AI systems within limited budgets.
Best Refurbished Lenovo ThinkSystem GPU Servers
Lenovo ThinkSystem SR670 V2 – AI & Deep Learning
Built for GPU-heavy work, the Lenovo ThinkSystem SR670 V2 supports up to 4 double-width GPUs such as NVIDIA A100, H100, or RTX 6000 Ada. Larger configurations can host up to 8 GPUs.
Feature | Specification |
GPU Support | Up to 4x A100/H100/RTX 6000 Ada |
CPU | Dual Intel Xeon Scalable (3rd Gen) |
Memory | Up to 2TB DDR4 |
Storage | Up to 8x NVMe or 16x 2.5" SAS/SATA |
Networking | 10GbE / 25GbE / optional 100GbE |
Use Cases | Deep learning, simulation, HPC |
Lenovo ThinkSystem SR650 V2 – AI & HPC
A balanced 2U platform for compute and GPU workloads. It supports up to 3 single- or dual-slot GPUs such as NVIDIA A40 or A30.
Feature | Specification |
GPU Support | Up to 3x A40, A30, RTX GPUs |
CPU | Dual Intel Xeon Scalable (3rd Gen) |
Memory | Up to 4TB DDR4 |
Storage | Up to 20x 2.5” or 12x 3.5” drives |
Networking | 1GbE / 10GbE / optional 100GbE |
Use Cases | AI inference, virtualization, HPC |
Lenovo ThinkSystem SE350 – Edge AI & GPU Acceleration
Compact and efficient for edge locations, the Lenovo ThinkSystem SE350 supports one NVIDIA T4 or A2 GPU and offers flexible power and connectivity.
Feature | Specification |
GPU Support | 1x NVIDIA T4 or A2 |
CPU | Intel Xeon D Series |
Memory | Up to 256GB DDR4 |
Storage | Up to 16TB NVMe SSD |
Networking | 1GbE, 10GbE, optional LTE |
Use Cases | Edge AI, IoT, analytics |
GPU Compatibility Summary - Refurbished Lenovo ThinkSystem GPU Servers
Lenovo ThinkSystem SR670 V2: NVIDIA A100, H100, L40, RTX 6000 Ada
Lenovo ThinkSystem SR650 V2: NVIDIA A40, A30, L40, RTX A6000
Lenovo Thinksystem SE350: NVIDIA T4, A2
(Use Lenovo GPU Enablement Kit for optimal driver and firmware support.)
Practical Considerations - Refurbished Lenovo ThinkSystem GPU Servers
1. Cooling and Airflow
High-end GPUs produce significant heat. Ensure proper airflow, fan upgrades, and heatsinks for 300W+ GPUs. The Lenovo ThinkSystem SR670 V2 benefits from optimized air ducts; third-party liquid cooling is optional but not standard.
2. Power Supplies
Lenovo ThinkSystem SR650 V2 → Dual 1600W recommended for multi-GPU loads
Lenovo ThinkSystem SR670 V2 → Dual 1600W or higher for A100/H100 configurations
Lenovo ThinkSystem SE350 → Optimized for lower power GPUs
Check PCIe 8-pin or 12-pin connectors for GPU compatibility.
3. PCIe Bandwidth
Lenovo ThinkSystem SE350: PCIe Gen3 (limited for top GPUs)
Lenovo ThinkSystem SR650 V2: PCIe Gen4 x16 slots (good for inference)
Lenovo ThinkSystem SR670 V2: Full PCIe x16 lanes per GPU (best for training)
4. CPU and Memory
Use Intel Xeon Gold or Platinum CPUs. At least 256GB RAM for moderate AI workloads, 512GB–1TB for deep learning. NVMe SSDs and RAID help handle large datasets.
5. Networking and Expansion
For distributed training, 100GbE or InfiniBand is ideal. SR670 V2 supports GPU-to-GPU NVLink bridges for faster data exchange
Lenovo Management and Software Support - Refurbished Lenovo ThinkSystem GPU Servers
Refurbished Lenovo servers retain XClarity Controller and XClarity Administrator functions for system monitoring and firmware management.Supported operating systems include Red Hat Enterprise Linux, Ubuntu, and VMware ESXi, all compatible with NVIDIA CUDA, TensorFlow, and PyTorch.
Buying Refurbished Lenovo ThinkSystem GPU Servers
Tested, cleaned, and certified servers
Refurbished GPUs with full performance testing
Up to 5-year warranty options
Pay after testing
Fast delivery with no long lead times
Refurbished Lenovo ThinkSystem GPU servers offer reliable, budget-friendly performance for AI and HPC workloads while keeping management tools, compatibility, and expandability intact.
Lenovo ThinkSystem Serves: Save Up To 80%
✔️ 5-Year Warranty – No Risk: Pay Only After Testing






Comments