4 min readJust now

–

Kyoto Region Update: Why scale out when you can scale up? A story of father-son bonding, structural stuffed animals, and unconventional GPU mounting.

Press enter or click to view image in full size

Where it all began. Building the first iteration (Ryzen 5 8500G) with my son. Little did we know this modest gaming PC would soon evolve into a headless monster.

1. The Origin: A Father-Son Project It all started with a simple goal: I wanted to build a gaming PC with my son. We went to the electronics store together, picked out the parts, and built it from scratch. It was a fantastic experience to teach him how computers work. At the time, it was a modest machine:

• CPU: Ryzen 5 8500G
• GPU: RTX 4060 Ti
• Case: A generic store-brand case (Dospara)

It was a great start, but as an infrastructure engineer, I couldn’t help but look at that machine when he wasn’t playing games and think… “What a waste of resources.”

2. The Awakening: Seeking Density I decided to dual-boot the machine with Ubuntu and started using it as a web server and for AI transcription tasks. I was immediately shocked. Even the modest Ryzen 5 8500G was significantly faster than the cloud instances I was paying for. That’s when the “Home Lab Virus” hit me. I didn’t just want a PC; I wanted a Monster. In Japan, we have a “Space Tax” — our apartments are small. So, my philosophy became: “Why add more servers when you can make one server do the work of three?”

Press enter or click to view image in full size

The Heart of the Beast. Upgrading to the Ryzen 9 9950X and 128GB of RAM. The massive Noctua NH-D15 G2 cooler dominates the motherboard, promising silence and performance.

I upgraded everything:

• CPU: AMD Ryzen 9 9950X (The Beast)
• Cooler: Noctua NH-D15 G2 (Silence is King)
• RAM: 128GB DDR5 (I wish I bought more…)
• Case: Fractal Design Define 7
• Power: Fractal Design Ion+ 2 Platinum 860W

Press enter or click to view image in full size

The Result?

• Ryzen 5 8500G Geekbench: Multi-Core ~7,600
• Ryzen 9 9950X Geekbench: Multi-Core 20,176

It was a 2.6x performance jump in the same physical footprint. This confirmed my theory: High Density > High Scale for home labs.

3. The Problem: The RTX 5080 is Too Big To complete the beast, I bought the new RTX 5080. But I hit a physical wall. The card is massive. I tried a standard vertical mount, but it was a disaster. The GPU choked against the glass panel, and the cables interfered with the massive Noctua CPU cooler. It was hot, loud, and ugly.

Press enter or click to view image in full size

The Failed Experiment. Attempting a standard vertical mount for the RTX 5080. It was too tight — the GPU backplate was practically hugging the CPU cooler, choking the airflow.

4. The Solution: The “Front-Mount” Hack (and the Pig) I looked at the empty space in the front of the Fractal Define 7 (usually reserved for HDDs or water cooling) and had a crazy idea. “What if I mount the GPU here, completely separate from the motherboard area?”

• The Mount: I removed the bottom covers and bolted the GPU vertically to the floor of the case.
• The Stability: I used L-brackets attached to the front fan rails to secure the top of the GPU.

Press enter or click to view image in full size

The “Front-Mount” Hack. I removed the bottom drive covers and bolted the GPU into the front chamber of the Fractal Define 7. Secured with L-brackets and connected via a long riser cable. No monitor output possible, but who needs a monitor for a server?

Press enter or click to view image in full size

• The Helper: During the installation, I needed something to hold the heavy GPU in place while I screwed it in. My partner’s stuffed pig became the most critical tool in the box. (See photo!)

Press enter or click to view image in full size

Advanced Structural Engineering. My partner’s stuffed pig provided critical structural support to hold the heavy RTX 5080 in place while I tightened the mounting screws.

The Caveat: This position blocks the HDMI/DisplayPort outputs. You literally cannot plug a monitor into it. But for me? I don’t care. This is a headless server for local LLM inference and transcoding. If I can SSH into it, I’m happy.

5. The Verdict Thanks to this layout and the “All Noctua” airflow:

Press enter or click to view image in full size

The Proof is in the Graphs. Despite the unconventional layout, the temperatures are ice cold. CPU stays under 80°C and GPU hovers around 30–50°C under load. Silence is golden.

Press enter or click to view image in full size

• CPU: 45°C — 80°C (under load)
• GPU: 30°C — 50°C
• NVMe: 20°C — 35°C

It is dead silent, incredibly powerful, and fits perfectly in my “Kyoto Region” lab. Sometimes, the best engineering solutions require a drill, some L-brackets, and a stuffed pig.