HYPERPC AMPERE RACK AI GPU Server | 2x RTX 5090 & Threadripper PRO

Client: Leading Technical University

Project: High-performance GPU server AMPERE RACK

Task: Provide resources for neural network training, mathematical modeling and Big Data work, eliminating dependence on cloud services

Problem

Standard laboratory PCs cannot handle modern workloads: training language models (LLM) takes weeks, and rendering complex physical simulations causes overheating and throttling. The university needed an autonomous solution capable of operating 24/7 under 100% load.

HYPERPC Solution

We designed and assembled a server in RACK form factor (4U) for installation in a server rack. The focus is on maximum computing density and fault tolerance.

Configuration and Selection Rationale

We selected components so that each part solves a specific engineering task:

1. Graphics Power: Tandem of 2x NVIDIA GeForce RTX 5090 GameRock

Why: This is the main striking force of the server. A pair of two flagship cards provides a total of 64 GB of video memory and over 43,000 CUDA cores.

Why it's perfect: For AI training (Deep Learning), VRAM volume is critical. 64 GB allows loading "heavy" neural network models entirely, without splitting them into parts, which speeds up training many times. This also ensures ultra-fast 8K rendering.

2. Computing Center: AMD Ryzen Threadripper PRO 7965WX (24 cores, 5.3 GHz)

Why: The professional WRX90 platform provides a huge number of PCIe lanes.

Why it's perfect: In regular PCs, graphics cards can "choke" due to low processor bandwidth. Threadripper guarantees that both RTX 5090s will operate at full power simultaneously. 24 high-frequency cores excel at linear calculations and code compilation.

3. Data Reliability: 256GB Samsung ECC DDR5

Why: RAM with error correction function (Error-Correcting Code).

Why it's perfect: In scientific calculations that last for days, even a single bit error can distort the final research result. ECC memory automatically finds and corrects errors, guaranteeing mathematical accuracy of simulations.

4. Cooling and Power System: Industrial Level

Custom Liquid Cooling + ARCTIC S12038-8K: We installed water blocks on the processor and both graphics cards, connecting them to an external radiator blown by 6 industrial fans at 8000 RPM.

Why it's perfect: At full load, the system generates almost 2 kW of heat. Standard cooling won't handle this here. Our solution maintains temperatures in the operating zone even during 24/7 rendering.

2000W Super Flower Platinum Power Supply: Provides stable current with power reserve, which is critical for the durability of expensive components.

Result

The university received a local supercomputer. Now students and graduate students can:

  • Train neural networks in hours, not days.
  • Conduct complex aerodynamic and physical simulations without failures.
  • Work with data arrays of hundreds of gigabytes in real time.

HYPERPC — creating tools for those who advance science forward.

Configuration

  • Graphics Cards:
    2x NVIDIA GeForce RTX 5090 GameRock
  • Processors:
    AMD Ryzen Threadripper PRO 7965WX [24 cores, 5.3 GHz]
  • RAM:
    256GB Samsung ECC DDR5
  • Cooling:
    Custom Liquid Cooling + ARCTIC S12038-8K
  • Power Supply:
    2000W Super Flower Platinum

We implement IT projects of any complexity

From a single server to building a private cloud. Contact us for detailed architecture development of your future system

Contact manager Other projects

Server Solutions

High-density rack systems for building IT infrastructure, HPC computing, virtualization and Big Data processing
To server catalog

Desktop Solutions

Graphics workstations (GPU Workstations) for resource-intensive tasks: neural network training (Deep Learning), 3D rendering and CAD/CAE design
To workstation catalog

Copyright ©2026 HYPERPC


main version