1억

Summary

In this instructional video, the team at YangCom Aquafarm embarks on assembling a high-performance server, incorporating four NVIDIA A100 80GB graphics cards, a setup that surpasses twenty million won in value. Emphasizing the choice of components similar to the client's current system for compatibility and stability, the assembly process showcases the installation of two Intel Xeon Gold 5220 CPUs, 332GB of DDR4 RAM from Samsung, and a 2TB SK Hynix Platinum P41 NVMe SSD. The video highlights the meticulous process of setting up the cooling system, utilizing industrial fans to address the significant noise generated by the fanless A100 cards, and concludes with the complex assembly of GPUs, CPU cooling blocks, and the final touches of cable management and coolant filling, drawing parallels to NVIDIA's DGX server assembly approach.

Takeaways

• 😊 The video shows assembling a high-performance server system with 4 NVIDIA A100 GPUs.
• 👨‍💻 The system uses dual Intel Xeon Gold 5220 CPUs for processing power.
• 💿 It has 8 sticks of 332GB Samsung DDR4 RAM, totalling 2.6TB memory.
• 🖥️ The GPUs are passively cooled for lower noise.
• 🔋 A 2000W power supply provides sufficient power.
• 🌡️ A custom liquid cooling system with radiators and pumps is installed.
• 🚿 The system uses PVC tubing in an in/out configuration for the cooling loop.
• 📦 High-end AlphaCool blocks and pumps are used for optimal cooling.
• ⏱️ Assembling the system is time consuming due to the complex cooling loop.
• 🎯 The end result is a GPU server similar to NVIDIA DGX systems.

Q & A

What is the purpose of assembling the high performance server?

The purpose is to assemble a high performance server for a client who does research, using specifications similar to their existing system for stability and optimization.

What GPUs are being used in the server?

4 Nvidia A100 GPUs without fans are being used for lower noise.

Why are the latest CPUs not being used?

To maintain compatibility and stability with the client's existing research systems and optimizations.

What steps are taken during the server assembly?

Key steps include installing the CPUs, memory, SSDs, GPUs, motherboard, power supply, radiators, tubing, and cooling system.

What cooling methods are used in the server?

Water cooling with radiators, pumps, reservoirs, tubing, and industrial fans.

Why are fanless GPUs used?

To reduce noise which was disrupting the client's work with previous servers.

What is the estimated cost of the server?

Over 200 million Korean won based on the 4 high-end GPUs costing over 20 million won each.

What tubing method is used?

Parallel in and out tubing between each GPU and CPU for optimal cooling flow.

How is the server similar to Nvidia DGX servers?

It uses a similar water cooling system and multiple high-end GPUs for research and computing.

What final steps are taken after assembly?

Filling the cooling system with fluid since the process is complex, and tidying cables.