4 NVIDIA A100 GPUs in a 1U Server 🔥 | Supermicro SYS-120GQ-TNRT
In this video, we take a deep dive into the Supermicro SYS-120GQ-TNRT, one of the most compact and densely packed 1U GPU servers ever designed for NVIDIA accelerators. Despite its small footprint, this platform can accommodate up to four full-size GPUs and is built for AI inference, large language models (LLMs), machine learning, HPC workloads, and rendering.
We explore every aspect of the system, including the Supermicro X12DGQ-R motherboard, cooling design, power delivery, and GPU installation process. You'll see how Supermicro engineers managed to fit four high-performance accelerators into a chassis that is only 1U tall.
In the second half of the video, we put the server to work. Equipped with four NVIDIA A100 80GB GPUs, we run a series of practical AI tests using the MiniMax 2.7 229B model. The goal is to evaluate real-world performance in LLM inference, code generation, data analysis, and infrastructure design tasks.
Today, GPU servers are no longer used exclusively for AI training. They are increasingly becoming the foundation for local LLM deployments, enterprise AI agents, inference clusters, and private AI infrastructure. Using the Supermicro SYS-120GQ-TNRT as an example, we explore what a 4× NVIDIA A100 platform can deliver and where such a system makes practical sense.
Timestamps:
00:00 - Introduction
00:40 - Unboxing
02:15 - First Look at the Server
03:00 - X12DGQ-R Motherboard
03:40 - Cooling System
03:52 - GPU Installation
04:59 - GPU Latch ASMR
05:51 - NVLink and AI Training
07:30 - Power Supplies
08:13 - 2x SFF NVMe Bay
08:59 - Installing CPUs and RAM
09:24 - Noise and Power Consumption
10:40 - Our Dedicated Cold Aisle
11:26 - SFBench Overview
11:52 - Choosing an AI Model
13:09 - AI Test #1: Financial Tracker
14:13 - AI Test #2: 3D Shooter Development
14:57 - AI Test #3: Interplanetary Mission Planning
15:42 - AI Test #4: Product Listing Generation
16:50 - AI Test #5: Data Center Design
18:58 - How Many Junior Engineers Can This Server Replace?
Our Telegram Channel: https://clck.ru/3RoQDu
SFBench: https://bench.serverflow.ru/
ServerFlow Blog: https://serverflow.ru/blog/