Building My Own PC Rig (Summer 2025)
In summer 2025 I finally built my own PC. The main goal was to support my thesis research in computer vision, without depending on the lab.
The lab has an AI server, but having access to it and being able to use it the moment I need it are two different things — it’s shared, so I can’t always jump on it right when I need to train something. On top of that, my laptop is from 2019 (i7 8th gen, RTX 2060 laptop edition), and it’s showing its age: wear and tear, plus performance that drops off once the workload gets heavy.
So I just built my own.
Here are the parts I bought, with the prices on the receipt at the time:
| Part | Model | Price (NTD) |
|---|---|---|
| CPU | AMD Ryzen 7 9800X3D | 15,700 |
| Motherboard | MSI MAG B850 Tomahawk MAX WIFI | 7,790 |
| Memory | Micron Crucial Pro OC 48GB x2 DDR5-5600 | 7,188 |
| Storage 1 | ADATA XPG GAMMIX S70 BLADE 2TB | 4,399 |
| Storage 2 | ADATA XPG GAMMIX S70 BLADE 2TB | 4,399 |
| Storage 3 | Acer Predator GM7 2TB | 3,890 |
| AIO Cooling | DEEPCOOL LT720 WH 360 | 3,290 |
| Case | Antec FLUX SE | 2,890 |
| PSU | Super Flower LEADEX VII 1300W Gold (White) | 6,890 |
| GPU | Gigabyte AORUS RTX 5090 MASTER 32GB | 99,990 |
| OS | Windows 11 Pro (Traditional Chinese) | 6,490 |
| Total | 162,916 |
What did it get me? On the fun side, I can run AAA games at high settings with no problem. But honestly the biggest win is the research, where the GPU does most of the heavy lifting. Training my models takes around 20GB of VRAM, and the RTX 5090’s 32GB handles that fine. If I’d gone with the cheaper 5080 and its 16GB, I’d have hit out-of-memory errors and not been able to train at all — so the “cheaper” card would have failed at the exact thing I built this PC for.
Worth every NTD.