Successful testing of AMPERE AI servers based on seven graphics accelerators

Engineering challenges and project issues

Designing a configuration with six or more flagship graphics cards at the RTX 5090 level is a complex technological challenge. During the development and testing process, HYPERPC specialists successfully resolved the following critical issues:

• Thermal energy removal: seven top graphics cards have extremely high TDP. At maximum load in computing tasks, they generate a volume of heat that requires an industrial approach to cooling.
• Stable power consumption: Peak loads from seven new generation GPUs and the Threadripper PRO processor require absolute power stability and a large safety margin.
• Physical placement and stability: GameRock series graphics cards are equipped with massive heatsinks, so engineers needed to integrate them into the dimensions of the 7U case and firmly fix each card.

Implemented HYPERPC engineering solutions

To ensure guaranteed uninterrupted operation 24/7, an integrated approach to server architecture was applied.

• Directional airflow system: a specialized 7U Rackmount case is used, divided into independent thermal zones. A block of high-speed server fans generates an air flow that blows through the radiators of graphics accelerators under high pressure.
• Enterprise Power redundant power subsystem: a server power system of the CRPS standard is integrated, consisting of four 3200W modules with a total power of 12,800 W. The architecture supports hot standby and allows calculations to continue if one module fails.
• Reinforced chassis design: A system of internal guides and supporting frames has been designed to firmly fix each graphics card and protect the PCI-E slots from mechanical damage.

Areas of application of AMPERE servers

The AMPERE line configurations (XR PLUS, XR PRO, XR MAX, XR ULTRA) are designed for the corporate sector, research institutes and professional studios.

• Machine learning and working with AI: training large language models (LLM), deep learning, neural network inference with high throughput thanks to the combined power of up to eight GPUs.
• Data Science and Predictive Analytics: processing of enormous amounts of data, complex mathematical and financial modeling.
• Professional 3D rendering: accelerating the rendering of complex scenes in film and animation, high-definition architectural rendering and real-time graphics.
• Scientific research: simulation of physical processes, molecular modeling, bioinformatics and hydrodynamic calculations, requiring continuous operation of equipment for days at maximum power.

Test results

The AMPERE XR PLUS configuration successfully passed all testing stages, including a continuous 72-hour stress test for maximum power consumption and parallel training of heavy machine learning models with 99% GPU resource utilization.

There was a complete absence of drop in clock frequencies on all graphics accelerators, stable voltage readings and absolute system fault tolerance. The HYPERPC AMPERE server line has confirmed its status as a reliable tool for implementing the most large-scale computing projects.