Defeating NVIDIA? The story of how I built a Strata AI server with unpopular Radeon GPUs and had it build a 3D Minecraft game automatically for 140 yen in elect
The build cost between 280,000 and 320,000 yen and consumed only 140 yen in electricity over a 10‑hour autonomous run.
Key points
- 10‑hour autonomous run produced 688,469 tokens for 140 yen electricity
- Strata engine enables AMD GPUs to match or exceed NVIDIA performance with 196,608‑token context
During the run, the autonomous coding agent Hermes Agent generated 688,469 tokens, completing a 3‑D Minecraft‑style game in a browser. The system achieved 75 to 80+ tokens per second at full power and 35–54 tok/s while running VRChat simultaneously.
The article compares the cost of this local setup to cloud APIs: Claude Sonnet 5.5 with cache usage costs about 2,100 yen, Claude Opus 5.5 about 4,250 yen, and GPT‑6 Astra about 10,600 yen, versus 140 yen for local electricity. It highlights that Radeon GPUs can match or exceed NVIDIA performance for large‑scale models when paired with the right software.
Model pages: Claude Sonnet 5.5 → · GPT-6.1 Sol → · Claude Opus 5.5 →
The story so far
4 episodes →- Defeating NVIDIA? The story of how I built a Strata AI server with unpopular Radeon GPUs and had it build a 3D Minecraft game automatically for 140 yen in electthis story
Defeating NVIDIA? The story of how I built a Strata AI server with unpopular Radeon GPUs and had it build a 3D Minecraft game automatically for 140 yen in electricity
note.com · 2 November 2026
Loading the full article…
This text was published by note.com and written by おぞ. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- IO Fund analyst sees $1T AI infrastructure opportunity for server firms · 1 src
- Innoscience shows GaN power solutions for AI servers at OCP Summit · 1 src
- Huawei reportedly plans to use Samsung, SK Hynix HBM in Ascend 960DT · 1 src
- Apple says foldable iPhone Duo sold out with supply timeline uncertain · 1 src
- Marvell agrees to buy Lumentum in $115 billion all-stock deal for AI connectivity · 3 src
Comments
via GitHub Discussions