High Performance Computing (HPC) utilizes technologies for large-scale, parallel computing. Modern HPC systems increasingly incorporate GPUs alongside traditional CPUs, often combining them within the same server for enhanced performance.

These HPC systems employ a dual root PCIe bus design to efficiently manage memory across numerous processors. This setup features two main processors each with its own memory zone, dividing the PCIe slots (often used for GPUs) between them for balanced access.

Key to this architecture are three types of fast data connections:

• Inter-GPU Connection: Uses NVlink for rapid GPU-to-GPU communication, allowing multiple GPUs to function as a single, powerful unit.
• Inter-Root Connection: Facilitates communication between the two processors via high-speed links like Intel's Ultra Path Interconnect (UPI).
• Network Connection: Employs fast network interfaces, typically Infiniband, for external communications.

This dual-root PCIe configuration optimizes both CPU and GPU memory usage, catering to applications that demand both parallel and sequential processing capabilities.