top of page
server-parts.eu

server-parts.eu Blog

Special Offer: New Supermicro SYS-420GP-TNAR+ with 8× NVIDIA A100 80GB — Only €55,000!

  • Writer: server-parts.eu server-parts.eu
    server-parts.eu server-parts.eu
  • 6 days ago
  • 3 min read

The Supermicro SYS-420GP-TNAR+ is a 4U rackmount GPU server designed for very demanding workloads such as AI training, deep learning, and HPC (High-Performance Computing). It is based on dual Intel Xeon Scalable CPUs and an NVIDIA HGX A100 platform with 8 GPUs.


Supermicro SYS420GP-TNAR+: Only €55,000

✔️ 3-Year Warranty – No Risk: Pay Only After Testing


Think of it as a smaller version of a supercomputer. It combines strong CPU performance, very fast GPU communication, and high-speed NVMe storage.


Supermicro SYS-420GP-TNAR+ 4U GPU server with 8× NVIDIA A100 80GB SXM4 GPUs, dual Intel Xeon CPUs, 512GB DDR4 memory, and NVMe storage for AI, deep learning, and HPC workloads. server-parts.eu. refurbished. used.


Hardware Overview: Supermicro SYS-420GP-TNAR+


Component

Details

Motherboard

X12DGO-6-P

GPUs

8× NVIDIA A100 80GB SXM4 (on NVIDIA Delta HGX baseboard)

CPUs

2× Intel Xeon Silver 4314 – 16 cores @ 2.4 GHz (32 cores total)

Memory

8× 64 GB DDR4-3200 RDIMM (512 GB total)

Storage

2× 1.9 TB NVMe PCIe 4.0 (2 DWPD) + 2× 960 GB NVMe PCIe 4.0 (1 DWPD, TCG Opal)

Networking

Intel X550 AIOM – 2× 10 GbE RJ45

Power Supply

4× 3000 W Titanium redundant PSUs

Condition / Warranty

New • 3 Years Supermicro Warranty • In Stock • 55 000 EUR (EXW)



System Architecture: Supermicro SYS-420GP-TNAR+


  • Dual Xeon CPUs: Each CPU handles up to 8 memory channels and communicates through Intel UPI. The Silver 4314 models give 32 cores total. For heavier AI or HPC jobs, the system can use Gold or Platinum CPUs (up to 40 cores / 270 W TDP).


  • Memory: 32 DIMM slots (16 per CPU) support up to 8 TB DDR4 RAM or 12 TB with Optane PMem 200. The installed Silver CPUs limit memory speed to 2667 MT/s (the modules are 3200 MT/s capable).


  • GPU Board: The NVIDIA HGX A100 baseboard connects 8x A100 80GB SXM4 GPUs with NVLink + NVSwitch. This gives each GPU ~600 GB/s connection to others, allowing them to act almost as one large GPU. Each A100 runs at up to 400 W and supports MIG (Multi-Instance GPU) – useful for splitting GPUs into smaller tasks.


  • PCIe & Expansion: 10 PCIe 4.0 x16 slots (5 per CPU). Supports extra network cards (25/100 GbE or InfiniBand), storage HBAs, or accelerator cards. The included AIOM slot carries the Intel 10 GbE NIC.


  • Storage: Front: 6 hot-swap 2.5″ bays (SATA/SAS/NVMe).2 × NVMe U.2 (1.9 TB each) for data, 2 × M.2 (960 GB each) for boot OS. RAID 0/1/5/10 possible for SATA drives via Intel RSTe.


  • Power and Cooling: 4 × 3000 W PSUs (2+2 redundancy) supply enough energy for 8 × 400 W GPUs and 2 CPUs. The system can draw over 3.5 kW under full load. 4 large fans and air shrouds keep GPU temperatures stable. Designed for data centers with good rack airflow and power provisioning.


  • Management: Integrated BMC (ASPEED AST2600) with IPMI 2.0 and remote KVM. Dedicated management port, plus front USB and status LEDs. Supports Supermicro SuperDoctor and Server Manager.


  • Chassis Design: 4U height (17.6 ″ W × 6.9 ″ H × 35.4 ″ D ≈ 900 mm depth). Modular two-drawer layout – top 1U for CPUs and memory, bottom 3U for GPUs and switch board. About 75 kg fully equipped. Rack rails included.



Performance and Typical Use: Supermicro SYS-420GP-TNAR+


This server is ideal for:


  • AI and Deep Learning Training — massive GPU performance with fast interconnect (NVSwitch).

  • HPC Workloads — simulations, CFD, genomics, and scientific modeling.

  • Data Analytics and Research — large memory and NVMe storage handle big datasets easily.

  • GPU Virtualization — using MIG to divide GPUs for multiple users or smaller jobs.


The A100 80GB GPUs deliver up to 20 × faster training speed than previous generation models. The dual-Xeon architecture ensures CPU bottlenecks are minimal for GPU tasks. Adding 100 GbE or InfiniBand adapters is common in multi-node clusters for faster data exchange.



Power and Cooling Requirements: Supermicro SYS-420GP-TNAR+


  • Max Load: ≈ 4 kW (when GPUs are fully active).

  • PSUs: 4× 3000 W redundant (80 Plus Titanium).

  • Cooling: Front-to-back airflow with high-speed fans.

  • Environment: Data center rack with stable temperature (10–35 °C) and sufficient airflow.

  • Noise: Very loud under load – rack use only.



Upgrade and Future Options: Supermicro SYS-420GP-TNAR+

Component

Upgrade Path

CPU

Supports all 3rd Gen Xeon Scalable CPUs (up to 40 cores, 270 W)

Memory

Up to 8 TB DDR4 or 12 TB with Optane PMem 200

Storage

6 front bays for additional NVMe/SATA/SAS drives

Networking

25 GbE / 100 GbE / InfiniBand cards supported

GPU Platform

Fixed to 8× A100 SXM4 HGX board – no GPU mixing possible

Software

Certified for NVIDIA CUDA, TensorFlow, PyTorch, and major HPC frameworks


This hardware will stay relevant for several years for AI, HPC, and research use. The A100 platform is still widely used in data centers and supports modern AI frameworks.



Final Details: Supermicro SYS-420GP-TNAR+

  • Price: 55 000 EUR (EXW)

  • Condition: New

  • Warranty: 3 Years from Supermicro

  • Availability: In Stock – Ready to Ship



Supermicro SYS420GP-TNAR+: Only €55,000

✔️ 3-Year Warranty – No Risk: Pay Only After Testing

bottom of page