Building an AI Server: DeepSeek Running Locally on H100 and Dual EPYC
We’re building an AI server powered by two AMD EPYC CPUs and an NVIDIA H100 GPU.
This workstation-class system is designed for deep-learning training and inference tasks. In this video, we show the complete build process - cooling setup, cable management, and component installation inside the case.
At the end, we run DeepSeek locally and check how this configuration performs with today’s large models.
The server is tuned for heavy AI workloads yet remains quiet, neat, and visually refined.
The configuration is ideal for large language models, local inference systems, and fine-tuning environments.
The build is based on the Gigabyte MZ73 server motherboard with dual AMD EPYC processors well-suited for inference workloads, and the NVIDIA H100 GPU — one of the most advanced accelerators for compute and neural-network tasks.
ServerFlow Telegram: https://t.me/serverflowofficial
ServerFlow Blog: https://serverflow.ru/blog/
Timestamps:
00:00 Intro
00:21 Motherboard
01:07 SSD
01:39 CPU
02:48 RAM
03:29 Coolers
04:59 Case
05:54 Power Supply
07:24 GPU NVIDIA H100
10:40 DeepSeek Locally
13:24 AIDA64 Memory Benchmark
13:49 GeekBench
13:56 Cinebench R23
14:06 Cinebench 2024
14:13 V-Ray Benchmark
14:21 Corona Benchmark