AI servers typically incorporate multiple accelerator cards such as GPUs and TPUs. These chips feature an enormous number of pins and extremely high signal transmission rates. Therefore, motherboards and accelerator cards require ultra-high-layer PCBs with 20 or even 30+ layers, along with HDI. The DGX A100 resembles a typical home computer and can be divided into five main hardware modules: Fan Module: Located at the front, the fan module consists of eight fans, which align with the standard 8U configuration found in traditional servers. Hard Drives: Positioned below the front fan. With six NVSwitch units on an A100-based system, the per-system value is RMB 1,170. High-Core CPUs Used to manage tasks and coordinate GPU workloads. Below, we round up the best GPU server configurations for your AI tasks. Most GPU servers have a CPU-based motherboard with GPU based modules/cards mounted on that motherboard. This setup lets you select. The Software Reference Architecture is comprised of individually optimized NVIDIA-Certified System servers that follow a prescriptive design pattern to ensure optimal performance when deployed in a cluster environment.
[PDF Version]