When it's comes to handling sensitive data and generating highly specific outputs, nowadays most of the company, lab or researchers, specially in like sectors medical, healthcare, finance, trying to develop their own custom made AI Agent or model, tailored to their unique requirements and build their AI infrastructures.
Inside PC Builders Bangladesh office, we often get offer or request to build others server, PC and list go on.
Last month, we got a request to build a Private AI PC (Infrastructure) to maintain their research work with their private data. So, the team buckle up and start this project. It looks simple but the collecting the Nvidia GPU is tough for logistics team. Total cost of this build $16,930 USD. Hence set-up cost a lot but in long-term, it gives lower operational cost to them.
Click here you can watch the entire build
So, what component use to build this PC and why?
we use AMD Ryzen Threadripper 7980X as processor. This is perfect for their job. 'Cause Threadripper 7980X offers top-tier multi-threaded and PCIe lane capacity, which is absolute for AI servers, especially those combining many GPUs. In budget, it offers performance per dollar compared to many Intel Xeon and AMD EPYC CPUs when it comes to AI and model training workloads.
In any AI Infrastructures, the most important thing is GPU, specifically NVIDIA GPU. NVIDIA GPUs considered essential assets for AI and machine learning tasks. In this build we use, NVIDIA RTX A6000, which is suitable option in my opinion, as it gives massive VRAM capacity, which very compatible to handle larger models (30B+ parameters with quantization), bigger datasets, or higher batch sizes, without memory limitations(what client needed). Also, ECC memory and workstation drivers offers Professional-Grade Reliability. want to use NVIDIA Blacwall GPU, but it still not available in consumer market. Maybe, primarily NVIDIA only offers the giant company, who pays well to them.
Then come to the Motherboard. In this section selected Asrock TRX50 WS, a good match for pairing with the Ryzen Threadripper and RTX A6000 GPU, as it offers up to 7 PCIe 4.0 x16 slots, critical for multi-GPU AI training setups.
In cooler section, try Abee Stem 360. I think it's an not bad choice,as it has special metal water cooling head known for its durability, also it good choice for maintaining the high thermal output of powerful components like the Threadripper 7980X.
In the entire build, the cheapest components is PC-Case. In my view as long as it offers excellent airflow to keep all hardware cooled effectively , it's OK.
At a glance of all components we used:
Processor: AMD Ryzen Threadripper 7980X
GPU: RTX A600
Motherboard: Asrock TRX50 WS
Ram: Samsung 32GB ECC Server Ram (great choice for this build)
Memory: WD BlackSN7100 2TB
PC Case: Corsair 3500X
Power Supply: Thermalright TG1350 1350W 80 Plus Gold Power Supply
Cooler: Abee Stem 360
Top comments (0)