• 
• 
• 
• 
• 
• 

Client: Engineering Center

Task: Radical acceleration of calculations on the "Virtual Proving Ground". Context: Existing systems could not handle complex computational fluid dynamics (CFD), creating downtime of several days for one calculation.

Problem Statement

The key pain point of the R&D division is time. Complex mathematical models and 3D meshes "hung" the old equipment. A solution was needed that could handle any complexity of models, and, critically, it had to be delivered in just 10 business days.

Solution: Custom Architecture for CFD

We abandoned boxed solutions in favor of a custom HYPERPC FORCE (Rack 4U) architecture, where every component works for parallel computing speed.

1. Computing Core: 256 AMD EPYC Cores

We installed two AMD EPYC 9754 processors.

Why this matters: Fluid dynamics perfectly parallelizes. 256 physical cores and 512 threads provide phenomenal computational density, reducing mesh rendering time by orders of magnitude.

2. Memory Subsystem: 1.5 TB Without "Bottlenecks"

24 modules of 64 GB each (DDR5-4800) were used.

Technical nuance: We utilized all 12 memory channels of each CPU. This provided maximum throughput to continuously "feed" data to 256 cores without a single idle cycle. The 1.5 TB capacity allows loading the heaviest mesh grids entirely into RAM.

3. Three-Tier Storage System

• NVMe SSD (3.84 TB): Works as a scratch disk for temporary calculation files — maximum IOPS speed is needed here.
• SATA SSD (RAID 1): Reliable storage for source files and final reports.

Result

The client received a system that reduced calculation time from several days to several hours. We proved that we can design, test, and deliver a world-class HPC solution in just 10 days, while A-brands are still approving specifications.