I built a private AI mini-cluster with Framework Desktop

Can we build a private AI cluster in a mini rack with Framework's Desktop motherboard?

Mentioned in this video:

- Framework Desktop: https://frame.work/desktop
- Level1Techs review of Framework Desktop: https://www.youtube.com/watch?v=L-xgMQ-7lW0
- Beowulf AI Cluster project: https://github.com/geerlingguy/beowulf-ai-cluster
- Blog post with all test results: https://www.jeffgeerling.com/blog/2025/i-clustered-four-framework-mainboards-test-huge-llms
- Other test results and benchmarks for the Framework Mainboard: https://github.com/geerlingguy/sbc-reviews/issues/80
- DeskPi RackMate T1 (Black - US): https://amzn.to/456iuXv
- DeskPi RackMate T1 (Black - EU): https://www.amazon.de/dp/B0FBFDZD4C?th=1

Framework sent the units used in this review, and DeskPi sent a mini rack and four rack trays to mount the Framework hardware. Neither company paid me, nor had any input into my testing or the production of the video. See my review video policies: https://github.com/geerlingguy/youtube?tab=readme-ov-file#review-videos

Support me on Patreon: https://www.patreon.com/geerlingguy
Sponsor me on GitHub: https://github.com/sponsors/geerlingguy
Merch: https://www.redshirtjeff.com
2nd Channel: https://www.youtube.com/@GeerlingEngineering
3rd Channel: https://www.youtube.com/@Level2Jeff

Contents:

00:00 - Desktop schmeshtop
00:32 - Nirav Patel explains the mini cluster
02:06 - Mainboard overview
04:16 - Single node performance
06:44 - AI and AMD mini-rant
07:47 - Single node AI benchmarks
09:03 - It's clusterin' time!
11:07 - A mini rack for the AI stack
14:27 - Large LLMs, clustered
16:14 - DeepSeek 671B and Llama 405B testing
18:02 - What about value?